<i>ROPPSA</i>: TV Program Recommendation Based on Personality and Social Awareness

Asabere, Nana Yaw; Acakpovi, Amevi

doi:https://doi.org/10.1155/2020/1971286

Mathematical Problems in Engineering

On this page

Abstract Introduction Discussion Data Availability Conflicts of Interest Supplementary Materials References Copyright Related Articles

Special Issue

Big Data Modelling of Engineering and Management

View this Special Issue

Research Article | Open Access

Volume 2020 | Article ID 1971286 | https://doi.org/10.1155/2020/1971286

ROPPSA: TV Program Recommendation Based on Personality and Social Awareness

Nana Yaw Asabere¹and Amevi Acakpovi¹

Guest Editor: Weilin Xiao

Received02 Feb 2020

Revised11 Apr 2020

Accepted16 May 2020

Published08 Jun 2020

Abstract

The rapid growth of mobile television (TV), smart TV, and Internet Protocol Television (IPTV) content due to the convergence of broadcasting and the Internet requires effective recommendation methods to select appropriate TV programs/channels. Many previous methods have been proposed to address this issue. However, imperative factors such as the utilization of personality traits and social properties to recommend programs for TV viewers remain a challenge. Consequently, in this paper, we propose a recommender algorithm called Recommendation of Programs via Personality and Social Awareness (ROPPSA) for TV viewers. ROPPSA utilizes normalization and folksonomy procedures to generate group recommendations for TV viewers who have common similarities in terms of personality traits and tie strength with a Target TV Viewer (TTV). Therefore, ROPPSA improves TV viewer cold-start and data sparsity situations by utilizing their personality traits and tie strengths. We conducted extensive experiments on a relevant dataset using standard evaluation metrics to substantiate our ROPPSA recommendation method. Results of our experimentation procedure depict the advantage, recommendation accuracy, and outperformance of ROPPSA in comparison with other contemporary methods in terms of precision, recall, f-measure (F1), and arithmetic mean (AM).

1. Introduction

Globally, recent decades have witnessed the emergence of novel challenges regarding television (TV) content consumption. TVs are machine screens that broadcast signals and convert them into multimedia pictures and sounds for educational and entertainment purposes. The evident convergence of Internet and broadcasting has paved the way for the proliferation of different TV technologies such as Internet Protocol TV (IPTV), mobile TV, and smart TV which can be viewed on multiscreen and device ecosystems [1–6]. In particular, the content of mobile TV is watched through mobile phones and smartphones, which makes them digitally ubiquitous.

Additionally, online social networking sites such as Facebook, Netflix, YouTube, and Twitter have complemented TV viewing experience by creating a large number of multiscreen applications [1–3]. Internationally, due to habit changes in accordance with diversification of TV content, active viewers experience difficult times deciding which programs to watch among thousands of channels as shown in Figure 1. Such situations create difficulty for TV producers to predict viewers’ interests and preferences in certain circumstances [1–6].

Consequently, the identification and selection of TV programs/channels have improved over the years. Electronic Program Guides (EPGs) have become available and more appropriate to use in comparison to historical and traditional methods such as browsing printed program guides and channel surfing [8–11]. Intelligent personalization techniques and systems can be innovatively utilized to significantly improve the effectiveness of searching and retrieving TV programs of interest.

Many researchers such as [1–6, 8–11] have utilized traditional recommender systems and techniques such as hybrid filtering (HF), collaborative filtering (CF), and content-based filtering (CBF) as shown in Table 1 to accomplish the task of searching, retrieving, and suggesting the right TV programs (items) for viewers through different recommender systems. Contextual information such as location, time, and physical conditions has also been applied in different recommendation scenarios [12].

Unfortunately, although the burden of information overload related to program consumption by TV viewers has been tackled to some extent, most of these existing personalized TV systems (e.g., [8–11]) did not exploit real-time social features/properties and personality traits of TV viewers to predict TV program interests/preferences. The incorporation of social (tie strength) and personality (traits) factors can help reduce the challenges of data sparsity (users rating a small proportion of items out of a larger number of available items) [13] and cold-start problems (new user and new item problems) [14] in TV recommender systems in order to significantly improve recommendation accuracy and quality of TV programs for viewers.

Consequently, this paper employs two main concepts, namely, personality and social properties/characteristics of TV users to generate effective and reliable group recommendations relating to TV program content. Social properties such as tie strength, closeness centrality, and degree centrality have been proved in literature [15–21] as very effective and important entities in social recommender systems. For instance, social ties and network centrality measures between attendees of a smart conference have positively been utilized to generate recommendations regarding conference attendees [15, 16], scholarly research papers [17, 18], and conference session venues [19–22].

Furthermore, some researchers who have worked on cold-start challenges in recommender systems [14, 23, 24] employed a wide variety of user profiles to generate recommendations without considering the personality of both new and existing users in different domains. Research has shown that personality is a stable and measurable psychological construct which can be used to explain various human behaviours using some traits [25]. This means that the concept of personality is capable of providing rich information. Moreover, Hu and Pu [26] proved that personality is the main and persistent factor which reflects the interests of people and predicts their behaviors. For example, the utilization of personality profiles can both provide recommendations for decision support and effectively solve cold-start issues [15, 16]. Therefore, it is appropriate to adopt new users’ personality as their profiles [15, 16, 25, 26] to carry out a deeper study on TV program recommendations for viewers.

In order to decipher the enumerated research problem above, we propose a recommender algorithm called Recommendation of Programs via Personality and Social Awareness (ROPPSA) for TV viewers. Specifically, ROPPSA utilizes normalization and folksonomy procedures to generate recommendations for TV viewers in groups that have strong similarities in terms of personality traits and tie strength with a Target TV Viewer (TTV). ROPPSA considers different sources of information including (i) personal (TV program interests of viewers), (ii) social (tie strengths between a TTV and TV viewers), and (iii) user personality traits between a TTV and TV viewers.

1.1. Contributions

The main contributions in this work are summarized as follows:(i)In our proposed recommendation method, we consider the social properties and personality traits of users by computing tie strengths and personality similarities between a TTV and TV viewers.(ii)Our recommendation method uses personality rating similarities coupled with a normalization procedure to generate reliable groups for personality group recommendations. Furthermore, our recommendation method utilizes the idea of tie strength computations and folksonomies to generate reliable groups for social group recommendations.(iii)The recommendation methodology developed in this paper reduces cold-start and data sparsity and is favourable for improving the social and personality awareness of smart and mobile TV viewers.(iv)Our proposed algorithm enables effective and reliable group recommendations of TV programs to TV viewers who have high tie strength and similar personalities with a TTV based on TV program interests.(v)Finally, our ROPPSA recommendation approach is evaluated and validated through an experimentation procedure, which utilized a relevant dataset for an experimental comparison with other contemporary recommendation techniques methods.

1.2. Structure of the Paper

The rest of this paper is organized as follows. Section 2 discusses related work. Section 3 describes our proposed ROPPSA recommender algorithm and model. Section 4 presents our experimentation procedure and results to show the effectiveness of our proposed ROPPSA method. Finally, Section 5 concludes this paper.

This section presents associated studies and literature relating to our study. We focus on related work pertaining to (i) personalized TV program recommender systems, (ii) social group TV recommender systems, and (iii) personality group recommender systems.

2.1. Personalized TV Recommender Systems

Globally, a lot of researchers have made considerable efforts to develop effective personalized TV program recommender systems. A brief report of such related studies is presented as follows.

Krstic and Bjelica [27] designed a personalized TV program guide that employs a one-class classifier tool with an autoencoder neural network. Similarly, Martinez et al. [9] proposed a personalized TV program recommender system called queveo.tv, which hybridizes collaborative filtering and content filtering.

Furthermore, Kim et al. [1] generated personalized TV program recommendations for a target user by utilizing CF to propose a TV program recommender algorithm. Through genre and similar channel preferences, a rank model was employed to generate TV program recommendations in [1]. Similar to [1], Pyo et al. [2] used sequential pattern mining such as viewing time and watching length of each TV program to develop an automatic recommender algorithm for personalized TV program scheduling.

Kwon and Hong [28] employed memory-based CF to propose a personalized program recommender algorithm for smart TVs. The proposed method in [28] improved recommendation performance of EPGs and smart TV recommender applications. Similar to [28], Song et al. [6] presented a novel solution for content recommendation through the customization of IPTV content in accordance with current user context and environment, thereby ensuring better user experience.

The fundamental concept of group recommendation tries to maximize the group satisfaction function, which is generally based on periodically repeating recommendations or a partial satisfaction function after the sequence of items [29]. Kim et al. [30] proposed a TV program recommendation method by merging multiple preferences. They used channels and genres of programs as features for their recommendation method. Similarly, Chaudhry et al. [31] considered a heterogeneous information network environment and proposed to extract the heterogeneous relationship information between the target viewer and different TV programs following different metapaths, to deliver good quality recommendation using the implicit feedback history data of the viewer.

2.2. Social Group TV Recommender Systems

Social TV is defined as a social media service whereby TV users share familiarities about TV programs in a social network that they are viewing [3, 32]. Effective social TV service is typically provided by considering two technicalities: (i) how to create social TV communities by grouping similar TV users? and (ii) how to recommend TV programs based on group and personal interests for personalizing TV? [3].

As a consequence of the above, Pyo et al. [3] employed two Latent Dirichlet Allocations (LDAs) and proposed a unified topic model, which recommends TV programs as a social TV service based on grouping similar TV users. Similar to [3], Wang et al. [33] established a group preference model by evaluating the external followees’ impact on group interest using external experts’ guidance for social TV group recommendation. Similar to [3, 33], Shin and Woo [34] proposed a socially aware TV program recommender for several viewers based on individual and group interests. In order to achieve this task, the authors merged user profiles by combining their common interests to generate social TV recommendations.

Shepstone et al. [35] presented a recommendation method which describes how audio analysis of age and gender in a group of TV viewers can be used to generate recommendations of TV programs for the group. Similarly, based on TV content and semantic reasoning techniques, Sotelo et al. [36] presented a method of TV program recommendation for groups of people.

2.3. Personality Group Recommender Systems

As a result of the fact that most methods model real-life conditions in group recommendation, the need for personality detection in a group is tremendously important [29]. In relation to personality inclusion in recommender systems, various types have been extensively proposed in [37–42].

In terms of personality group recommendations, Quijano-Sachnez et al. [43] described some new concepts of enhancing recommendations to people in different groups. Their approach maximized global group satisfaction by considering social associations and personality among each group. Similarly, Quijano-Sachnez et al. [44] ensured agreements within a group by defining alliances through personality and trust.

Similar to [43, 44], Quijano-Sachnez et al. [45] utilized three imperative features: past memory, trust, and personality to generate group recommendations. Similar to [45], Quijano-Sachnez et al. [46] proposed an entrusting-based recommendation method, which includes an analysis of personality in conflict situations and group characteristics such as size, structure, and trust between group members.

Recio-Garcia et al. [47] presented an innovative method of generating recommendations to groups based on CF and group personality composition. Similarly, through the utilization of personality and trust, Recio-Garcia et al. [48] proposed a Decision Support System for groups of people, where each user entrusts an agent that represents his/her interests. In order to generate personality group recommendations, Prada et al. [49] introduced diversity in the behaviour of social agents through a computational model of personality based on the Five-Factor Model (FFM).

In comparison with the existing TV program recommendation methods described above, our proposed ROPPSA method exhibits novelty and advantage of integrating social and personality awareness through the respective computations of tie strengths and personality similarities of TV users (TTV and other TV viewers) in order to establish strong ties and highly similar personalities between them for effective TV program recommendations.

Our method paves the way for TV users with strong ties (social relationships) and similar personalities to be grouped so that similar TV viewer groups are recommended similar TV programs as a result of their relationships and correspondences with one another. Consequently, personalized TV program recommendations are possible with the similar TV user groups related to social and personality awareness.

The fundamental recommendation process of ROPPSA is illustrated in Figure 2. Figure 2 involves TV viewers, various recommendation entities, and personality and social group recommendations of TV programs. The integration of individual user profiles represents a possible solution for recommending TV programs to multiple viewers. This process is referred to as group modeling or group recommendation [31].

To the best of our knowledge and with reference to the literature above, this is the first time tie strength and personality traits are being utilized to generate group recommendations for TV viewers.

In the next section, we present details on the design of our proposed ROPPSA recommendation method for TV viewers/users. Specifically, personality traits for TV recommendation are not available in the literature. This reason further corroborates our innovative objectives for proposing ROPPSA as a recommendation method for TV viewers in our society.

3. Algorithmic Design of ROPPSA

In this section, we present the fundamental concept and architecture of ROPPSA. Figure 3 illustrates that our ROPPSA recommendation model generates both social group and personality group TV program recommendations through initial computations of tie strengths and personality similarities among mobile TV viewers, respectively.

Referring to Figure 3, the TV Viewer Crawler (TVC) collects and sends personality trait ratings of individual TV viewers to the Group Profile Organizer (GPO). The GPO is responsible for creating group profiles in accordance with personality trait ratings and interests of participants using a normalization procedure. The Similarity Profiler (SP) computes similarities of personality trait ratings between the Target TV Viewer (TTV) and the created personality group profiles (pgp_b) in order to generate reliable and efficient personality group recommendations of TV programs based on the interests of the TTV.

The Tie Strength Manager (TSM) computes the contact frequencies and durations between the TTV and other TV viewers so that in accordance with their tie strengths, the Folksonomy Group Profiler (FGP) allocates groups for wider coverages of reliable and efficient social group recommendations based on program interests of the TTV.

The program interests of smart and mobile TV viewers can vary at any time as a result of modifications in TV schedules of a particular broadcasting station. Due to such variations, the recommendation process in ROPPSA depends on both mobile and smart TV viewer profiles in the current viewer situation. Due to the fact that ROPPSA operates on smart and mobile TVs, ROPPSA requires standard mobile and smart TVs with integrated operating systems, Internet, and web 2.0 features which use a web browser as a client supported by a programming language. We elaborate further on our ROPPSA recommendation framework below.

3.1. Capturing TV Program Preferences in ROPPSA

As stated earlier, in order to determine TV program preferences of TV viewers, ROPPSA initially utilizes and computes similar personalities between a TTV and other (grouped) TV viewers for its personality group recommendation service. Consequently, ROPPSA employs an explicit approach that allows TV viewers to input specific personality trait ratings from 1 to 5 using the Big Five Personality Traits (BFPT) [25]. BFPT consists of the following:(i)Extraversion: talkative, outgoing, energetic assertive, etc.(ii)Agreeableness: trusting, sympathetic, forgiving, generous, kind, appreciative, etc.(iii)Openness: wide interests, innovative insightful, imaginative, curious, artistic, etc.(iv)Conscientiousness: responsible, thorough, reliable, planful, organized, efficient, etc.(v)Neuroticism (emotional stability): worrying, unstable, touchy, tense, self-pitying, anxious, etc.

ROPPSA further employs and computes tie strengths between a TTV and other (grouped) TV viewers (using their contact durations and frequencies) for its social group recommendation service.

3.2. Personality Group Profiling

Recommendation to groups usually involves the application of precise and appropriate characteristics in order to develop a compromise for different group semantics that validate disagreements and agreements among users [31, 44]. We organize our personality group profiling process illustrated in Figure 4 by adopting the arrangement of group recommendation in CF presented by Bobadilla et al. [7] and Asabere et al. [17]. Figure 4 demonstrates how our recommendation method combines TV viewer data and obtains data involving groups of TV viewers in four different steps.

The personality profiles of users (TV viewers) define their interests in terms of TV programs. Consequently, group profiles define common features used by individual TV viewers. We conformed to group profiling of individual TV viewers, by adopting the socially aware TV program recommender method used in [34]. As a consequence, we initially combine TV viewer profiles based on their personality trait ratings. The personality trait ratings of TV viewers are further arranged into groups by the GPO due to the fact that they have different traits and ratings.

As shown in Figure 4, the rating matrix of personality trait ratings by TV viewers is verified by TV_c in step 1. In steps 2 and 3, GPO generates neighbourhood intersections and group prediction phase (normalization), based on similar personality traits rated by TV viewers using (1). In the last step, SP decides which TV programs to recommend and generates a personality group recommendation based on the personality similarity between a TTV and the created personality group profiles (pgp_b).

Consequently, it is imperative to compute the weighted average associated with each personality trait rating using (2). This computation describes the program preferences of the TV viewers in order for the GPO to create a preference model for generating each group profile. As shown in Figure 4, GPO utilizes the K-nearest neighbour methodology by identifying unique personality trait ratings of TV viewers and categorizing them as a group [17, 31, 34]. Therefore, TV viewers with similar personality trait ratings are classified as nearest neighbours and grouped together.

Personality trait ratings of TV viewers are done on a scale of 1–5, where 1 and 5 are the lowest and the highest preference ratings, respectively. Let the TV program preferences of each group profile be PGP = {pgp₁, pgp₂, pgp₃, ⋯, pgp_b}. The normalization process by GPO interconnects common neighbours and predicts different groups of TV viewers [17, 18]:

We employ (1) to compute the normalized (n) values of personality trait ratings which have been rated differently by different TV viewers. In (1), TV_a,b denotes individual personality trait ratings of the TV viewers, for a particular personality trait, where a is the notation for a particular TV viewer and b is his/her rating for a particular personality trait. The precise maximum and minimum ratings of the TV viewers for a particular personality trait are, respectively, represented as TV_max and TV_min. In addition, PR_max and PR_min represent the overall highest and lowest personality trait ratings in the experimental dataset. It must be emphasized that, in (1), PR_max and PR_min are, respectively, equal to 5 and 1, representing the highest and lowest possible personality trait ratings in the dataset [17, 18]:

In (2), the values attained from (1) are totaled and divided by the number of TV viewers (NTV) who allotted ratings to a particular personality trait. Consequently, through (2), we compute and attain an average normalized value which designates the creation and allocation of a group profile for particular TV viewers. The TV viewers are allocated into their respective group profiles if their rating for a particular personality trait is greater than or equal to the corresponding average normalized value (pgp_b) of their group [17, 18].

With reference to Table 2, we demonstrate an example of our group profiling and normalization technique. In Table 2, there are four TV viewers with different personality trait ratings. For instance, in order to generate a group profile for the personality trait “Openness,” the first step of our proposed recommender algorithm is to count the number of available ratings related to “Openness.” Table 2 shows that there are four different ratings associated with “Openness” rated by each of the four TV viewers.

The normalized value of TV_a’s annotation on “Openness” with a rating of 3 is computed as follows:

Similarly, the normalized value of TV_b’s annotation on “Openness” with a rating of 2 is computed as follows:

Furthermore, the normalized value of TV_c’s annotation on Openness to experience with a rating of 4 is computed as follows:

Additionally, the normalized value of TV_d’s annotation on Openness to experience with a rating of 2 is computed as follows:

Consequently, the creation of a personality group profile (pgp_b) for the personality trait “Openness” in Table 2 is computed as follows:

The above normalization and personality group profiling approach in our proposed ROPPSA recommendation model illustrates that allocation of TV viewers to a personality group profile associated with “Openness” ratings should be assigned to TV viewers with ratings of 3.4 or above. This implies that, as per the data on Table 2, only two TV viewers, namely, TV_a and TV_c, can be assigned to that particular personality group profile for personality group recommendation of TV programs.

We employ equation (8) to determine the Personality Similarity (PSim) Score between TTV and pgp_b, i.e., PSim (TTV, pgp_b). In (3), TTV and pgp_b are represented by a and b, respectively. The average of all personality trait ratings of a and b is symbolized by and , respectively. Additionally, the personality trait ratings of a and b with one of the personality traits x are denoted by P_a,x and P_b,x, respectively [15, 16, 26]:

Our algorithm utilizes and returns a PSim Score between −1 and 1, where 1 indicates that TTV and pgp_tv have precisely the same or quite similar ratings and −1 denotes otherwise [15, 16, 26]. As a result of the fact that our recommender algorithm generates multiple personality group profiles for recommendation of TV programs, it loops through different TTV and pgp_b transactions in accordance with the dataset using (8). Furthermore, with reference to the range for PSim Scores during experimentation, we set a threshold α in (9) for PSim computations between TTV and pgp_b as follows:

3.3. Tie Strength

In a social network, tie strength is defined as the extent to which users are close to each other in terms of social/collaborative relations. A high tie strength normally depicts strong social relations such as friendship and familiarities among users in a social network. Users in a social network with low tie strengths have weaker relationships in terms of closeness. In order to acquire essential and important information from users in a social network, the concept of ties represents a major channel for collaboration and communication. Ties establish communication links in a social network, thereby improving and providing key elements for social cooperation and collaboration among users [15–21].

The concept of social ties/tie strength has been applied in different socially aware recommender algorithms and techniques [15–21]. Available evidence in the literature shows that these algorithms have performed considerably well in the alleviation of issues regarding cold-start and data sparsity as well as improved social recommendation accuracy [15–21]. Using the following equation, we measure and evaluate the tie strength between TTV and another TV viewer, TV_a

In (10), f_TTV,TVa is the social contact frequency between TTV and TV_a in the time frame T and d_TTV,TVa is their social contact duration within the same time frame T.

3.4. Social Group-Tie Profile Generation

The expansion of different social network communities substantiates innovative procedures and arrangements regarding the important role of self-organization principles. In the scope of this paper, we are interested in an algorithm which involves sharing of social resources, through the usage of a knowledge demonstration process called folksonomy [17, 50].

In order to generate social tie-group profiles in our ROPPSA recommendation model, we initially utilize the concept of folksonomies by connecting a TTV to other TV viewers in the dataset. In this paper, a folksonomy can further be described as a hypergraph G = [TTV, TV, H], where TTV represents Target TV Viewers, TV represents other TV viewers, and H = [h₁, h₂, h₃, ⋯, h_n] denotes the set of hyperedges which only exist among the nodes (TV viewers) in different sets [17, 50].

With reference to Figure 5, a TTV is connected to different TV viewers through hyperedges h₁ and h₂, respectively. TV viewers TV_a, TV_b, TV_c, and TV_d are connected to the TTV through hyperedge h₁. Furthermore, TV viewers TV_e, TV_f, , and TV_h are connected to the TTV through hyperedge h₂. We employ (10) to compute the tie strengths between TTV and all other TV viewers connected to hyperedges h₁ and h₂. After computing the tie strengths, we utilize (11) to compute a threshold for the social tie-group profile (stgp_tv) which is connected through hyperedge h₃:

Using a time frame (T) of 660 minutes, we present an example of our social tie-group profile generation process in Table 3 and further explain how (10) and (11) are applied to generate social group recommendations for TV viewers.

Table 3 depicts the computation of tie strengths between TTV and other TV viewers using (10) which applies respective contact durations and frequencies. A close view of Table 3 depicts the tie strength for each computation of the TTV and other TV viewers. We employ (11) to compute the threshold of the social tie-group profile for the tie strength displayed in Table 3 as follows:

The above social tie-group profiling approach in our proposed ROPPSA recommendation model demonstrates that the allocation of TV viewers to a social tie-group profile based on tie strength should be allotted to TV viewers who have a tie strength of 0.3 or above with a TTV. This implies that, as per the data on Table 3, only four TV viewers, namely, TV_a, TV_b, TV_c, and , can be assigned in that particular social tie-group profile to generate a social group recommendation of TV programs. This is depicted in the connections of hyperedge h₃ in Figure 5.

3.5. ROPPSA Recommender Algorithm

Our proposed ROPPSA recommender algorithm (Algorithm 1) displays the required inputs and outputs for personality and social group recommendation procedures. Variables in our recommender algorithm are declared in steps 2–4. In step 5, unique personality groups are identified. Steps 6–12 compute normalized personality groups using (1) and (2) and allocate TV viewers to their corresponding groups. Using (8), steps 13–16 compute the personality trait similarity between TTV’s and personality group profiles (pgp_b) in order to recommend TV programs to personality group profiles based on personality trait similarity threshold utilized through (9).

	Input:
	Target TV Viewer (TTV)
	TV viewers (TV_s)
	Personality trait ratings of TTV and TV_s
	Tie strengths of TTV and TV_s
	Output:
	TTV personality and social group recommendations for TV_s
(1)	procedure Personality and Social (Group) Recommendation of TV programs
(2)	//Declare and Initialize Variables
(3)	a and b //integer variables
(4)	personality_group_threshold, PSim_threshold and tie_strength_group_thresh; //floating variables
(5)	Identify unique personality trait ratings and classify as group
(6)	//Compute normalized values of personality groups using (1) and (2)
(7)	//Allocate TV viewers to respective groups
(8)	for every TV viewer do
(9)	if (TV viewer[a].Rating ≥ personality_group_threshold) then
(10)	Allocate TV viewer[a] in pgp[b];
(11)	end if
(12)	end for
(13)	//Compute PSim of TTV and pgp_b using (8) and (9)
(14)	if (PSsim (TTV, pgp_b) ≥ PSim_threshold then
(15)	recommend TTV’s TV programs to pgp_b;
(16)	end if
(17)	for every TV viewer do
(18)	Compute tie_strength between TTV and TV_susing (10) and (11)
(19)	if (Tie_Strength (TTV, TV_s) ≥ tie_strength_group_thresh) then
(20)	Allocate TV viewer[a] in stgp[b] and recommend
	TTV’s TV programs to stgp_b;
(21)	end if
(22)	end for
(23)	end procedure

Similarly, steps 18–22 compute the tie strength between TTVs and TV viewers using (10) and allocate TV viewers into social tie-group profiles using (11) as a threshold for social group recommendation of TV programs.

4. ROPPSA Experimentation Procedure

This section demonstrates our experimentation procedure for the evaluation of ROPPSA. Using an extended version of the ATU dataset utilized in [21], we present our experimental results in comparison with other existing and relevant methods.

4.1. Experimental Dataset

In order to substantiate the procedure of our proposed recommender algorithm (ROPPSA), it was imperative that we utilize precise and relevant data for our experimentation process. Therefore, as stated above, we utilized and extended version of the Accra Technical University (ATU) dataset published in [21]. The ATU dataset in [21] needed to be extended because it did not contain personality trait ratings required for the experimentation in this paper. Consequently, we gathered more user, tie strength, and personality trait data from ATU students which we represented as TV viewers. Data were gathered from students in Higher National Diploma (HND) Marketing (n = 2864), HND Computer Science (n = 720), and HND Building Technology (n = 812).

Therefore, the ATU dataset utilized in this paper contains a total of 4396 TV viewers (2421 males representing 55.07% and 1975 females representing 44.93%). Tie strength details of the ATU dataset utilized in this paper are illustrated in Figures 6 and 7.

The personality data gathered in our ATU dataset involved personality trait ratings by the ATU students (designated as TV viewers) in the dataset. Among the BFPT, students rated each personality traits using a scale of 1 to 5. The total number of personality ratings for all the traits combined in the ATU dataset is 22,541. Figure 8 illustrates the personality trait data utilized in our ATU dataset for experimentation.

Additionally, in order to ensure and substantiate recommendation accuracy in our experimentation procedure, we gathered TV program interests of TV viewers in accordance with the following: movie/TV series—Horror, Action, Comedy, Adventure, and Drama; sports/games—Football, Basketball, Hockey, Boxing, and Athletics. Tables 4 and 5 illustrate data regarding TV program interests of TV viewers in the ATU dataset in accordance with their ratings.

In relation to the ATU dataset, the total time frame (T) which is 1140 minutes utilized in [21] remained the same. Consequently, using (10), we computed Tie_Strength_TTV,TVa(t) = (80 × 7)/1440 and achieved a result of 0.5 as the maximum (highest) Social Contact Score (SCS) between a TTV and other TV viewers for positive and effective social recommendations [15–21]. In ROPSAA, the quality of social group recommendations for TV viewers is determined through the corresponding level of tie strength computation a TTV has with a social tie group.

Through the utilization of (11) in accordance with the ATU dataset, more effective social group recommendations in terms of quality and accuracy were generated with SCSs between 0.3 and 0.5. Therefore, the priority TV program recommendation list for a particular social contact group was established based on computed recommendation values that fell within this threshold. However, SCSs which fell out of the above threshold were considered as unsuccessful TV program recommendations. Consequently, we experimentally established 0.1 ≤ SCS ≤ 0.5 as the range for TV program recommendation based on social contact (relations) and assigned a social group recommendation threshold of 0.3–0.5 in accordance with the dataset.

Typically, the computation of Rating Similarity Scores (RSSs) between two users/products is between −1 and 1. A value of 0 indicates that there is no association between two users or variables [51]. Consequently, using (8), we experimentally established 0.6 ≤ RSS ≤ 1 as the range for TV program recommendation based on personality traits and assigned a personality group recommendation threshold of 0.8–1 using (9) according to the ATU dataset for successful recommendations in terms of prediction accuracy. Conversely, computed RSSs which fell out of the above threshold were considered as unsuccessful TV program recommendations.

4.2. Experimentation Setup and Evaluation Metrics

In our experimentation setup, we utilize a cross-validation technique to evaluate ROPPSA by initially partitioning our ATU dataset in training (80%) and test (20%) sets. The evaluation metrics employed in our experimentation of ROPPSA in the case of personality group recommendations focused on prediction accuracy. Therefore, we utilized root mean square error (RMSE), mean absolute error (MAE), and normalized mean absolute error (NMAE) in this regard. Lower values of RMSE, MAE, and NMAE corroborate better performance. RSME is defined as follows [7]:

Additionally, MAE is defined as follows:where denotes the number of personality trait ratings in the test set. and , respectively, represent the real and prediction values of the personality rating in . As a result of the fact that different recommender systems/algorithms may use different numerical scales, we employed NMAE in our experiments to ensure that experimental errors can be fully expressed on a normalized scale. We therefore used (15) to compute NMAE. In (15), PR_max and PR_min are denoted as the upper and lower limits of user personality trait ratings in our ATU dataset, respectively. Consequently, in accordance with our ATU dataset, PR_max = 5 and PR_min = 1 [7]:

In relation to social group recommendations, we focused on recommendation quality and therefore employed accuracy, recall, and f-measure (F1). Higher values of accuracy (A), recall (R), and F1 substantiate better performance. We employed (16) to compute the preciseness of a recommender algorithm by utilizing the accuracy metric as follows [7]:

Additionally, we utilized the recall metric to measure coverage of all good and successful TV programs using the following equation [7]:

In (16) and (17), Num (N, tv) represents the number of relevant TV programs in the TV recommendation list for social group recommendation, Num (TV) in (16) denotes the number of all TV program recommendations and Num (tv) in (17) denotes TV programs that are relevant to a particular social group. Furthermore, in order to substantiate the results obtained through (16) and (17), we computed F1 using the following equation [7]:

4.2.1. Baseline Methods for Comparison

As a result of the fact that ROPPSA generates two main types of recommendations, namely, personality and social group recommendations, it was imperative to employ relevant methods for experimental comparison. Consequently, in order to validate our recommendation results in this paper, we compare ROPPSA to the following relevant methods described below.

In relation to personality group recommendations for TV programs, we utilized the work done by Kim et al. [30] and Chaudhry et al. [31] and represented them as T1 and T2, respectively, for comparison in our experiments. Both T1 and T2 are relevant personalized group TV recommendation methods which involved similar approaches of combining/merging individual user profiles/models to select TV programs to suit groups of TV viewers.

Furthermore, in relation to social group recommendations for TV programs, we exploited the work presented by Shin and Woo [34] and Wang et al. [33]. We represented them as T3 and T4, respectively, for comparison in our experiments. Both T3 and T4 are relevant social group TV recommendation methods which involved similar approaches regarding the establishment of group preference profiles/models from individual TV viewer profiles which relate to common social similarity and interests in terms of TV programs. The main goal of our experimentation procedure was to verify the effectiveness of ROPPSA and improve recommendation accuracy by alleviating issues of data sparsity and cold-start in TV program recommendation. The experimental results below present evidence on how the above goals were achieved.

4.3. Experimental Results and Analysis

In order to evaluate the experimental results effectively, we validated our experimentation procedure by answering the following questions:(i)In comparison with T1, T2, T3, and T4, what was the overall assessment of ROPPSA in terms of TV program recommendation?(ii)During our experimentation procedure, was there a reduction of cold-start and data sparsity in ROPPSA, compared to T1 and T2 for personality TV program group recommendations and T3 and T4 for social TV program group recommendations?

In terms of personality group recommendations, we utilized RMSE, MAE, and NMAE to evaluate ROPPSA based on the computations and measurements of similar personalities between a TTV and groups of TV viewers as described above.

As shown in Figure 9, at the maximum Rating Similarity Score (RSS) of 1.0, ROPPSA attained the lowest MAE (0.763) in comparison with that of T2 (0.824) and T1 (0.937). Similarly, in Figure 10, at the maximum RSS of 1.0, ROPPSA achieved the lowest NMAE (0.190) in comparison with T2 (0.205) and T1 (0.234). Furthermore, the RMSE value of ROPPSA in Figure 11 is 0.759 at an RSS of 1.0, in comparison with T2 (0.820) and T1 (0.933). Subsequent results of MAE, MMAE, and RMSE in Figures 9–11 corroborate the effectiveness of ROPPSA.

Table 6 summarizes the performance comparisons of the above recommendation methods on the ATU dataset in terms of MAE and RMSE, respectively. The experimental results show that ROPPSA outperforms T1 and T2 in terms of MAE and RMSE. This is because the use of personality trait information improves recommendation accuracy and performance by compelling TV viewers to be more similar in terms of personality. It can also be observed that T2 and ROPPSA outperform T1 in terms of RMSE and MAE due to the same utilization of personality information. This further proves and substantiates the necessity and effectiveness of integrating personality information into TV program recommendation methods for performance improvement.

The experimental results in Table 6 and Figures 9–11 reveal that the use of personality as a TV program recommendation entity for TV viewers can help to further improve prediction accuracy by determining more similar group TV viewers because direct personality information expresses similarities between TV viewers’ program interests to some extent.

In our experimentation analysis, Figure 12 shows that at the maximum threshold value for social contact (0.5), ROPPSA achieved the highest accuracy value (0.237) when compared with T3 (0.177) and T4 (0.062). ROPPSA therefore generated more effective and successful TV program recommendations (high accuracy) than T3 and T4.

Higher recall specifies that a recommender algorithm returned the most relevant and effective results (TV recommendations) and exhibited minimal false-negative errors. According to Figure 13, the recall values of ROPPSA are higher than those of T3 and T4. Specifically, at the highest Social Contact Score of 0.5, ROPPSA achieved a higher recall (0.778) than T3 (0.569) and T4 (0.515). Therefore, ROPPSA outperformed T3 and T4 in terms of recall and hence generated more relevant and successful social group recommendation of TV programs (high recall).

In order to verify the transparency and authenticity of our experimental results for accuracy and recall in Figures 12 and 13, we computed F1 using (18). A critical view of Figure 14 shows that ROPPSA attained higher results (0.363) in terms of F1 in comparison with T3 (0.269) and T4 (0.112). This correspondence is due to the efficacy of information retrieval with respect to how ROPPSA devoted more importance to both accuracy and recall in terms of social group TV program recommendation. Furthermore, the positive transparency of F1 in social group recommendations is due to the outperformance of ROPPSA in accuracy and recall evaluation metrics in comparison with T3 and T4.

Additionally, in relation to the arithmetic mean (AM) results of accuracy and recall using (19), ROPPSA achieved higher results in comparison with T3 and T4. As shown in Table 7 and Figure 15, in terms of social group recommendation, at the highest tie strength (0.5), we attained AM results of 0.507, 0.373, and 0.289 for ROPPSA, T3, and T4, respectively:

Our approach proves that the AM results attained for ROPPSA were the highest in comparison with the F1 (harmonic mean) results, which accordingly corroborates that AM should always be higher than harmonic mean regarding the retrieval effectiveness of a recommender system/algorithm [52].

4.4. Summary of Experimental Results

In summary, it can be verified from Figures 9–15, as well as Tables 6 and 7, that ROPPSA reliably attained more favourable results in all the utilized evaluation metrics. This observation indicates that our approach is more suitable and robust due to its capability to make use of tie strengths and personality trait ratings of TV viewers. Consequently, in comparison with the baseline methods, ROPPSA alleviated the challenges of cold-start and data sparsity due to the unlimited utilization of both personality and social features of TV viewers.

Our proposed ROPPSA recommender algorithm outperformed both T1 and T2 as well as T3 and T4 in terms of personality and social group recommendations, respectively. This exemplifies the importance of personality group profile normalization and social group relationships/influences in generating effective and efficient TV program recommendations for TV viewers.

5. Discussion and Concluding Remarks

With reference to our research findings and our proposed recommendation method (ROPPSA), it is extremely important and vital to note that, globally, the data protection law of every nation has an objective to protect individuals’ fundamental rights to privacy and the protection of their personal data [53, 54]. Ghiglieri et al. [54] undertook an online survey which involved 200 participants to determine whether consumers were aware of smart TV-related privacy risks. The main findings in Ghiglieri et al. [54] depicted that generally participants are unwilling to disconnect the Internet from their smart TVs because they value the smart TV’s Internet functionality more than their privacy [54].

In Germany, a sectoral investigation was upgraded into a concrete policy guidance for smart TV stakeholder/users in relation to privacy/data issues of television services in the nation’s ecosystem. Additionally, the Netherlands has recently witnessed the implementation/enactment of three enforcement actions by the Dutch Data Protection Authority against leading providers of interactive television services [53]. These examples corroborate that the effective implementation of our proposed ROPPSA method should be done in cognizance with the Data Protection Act/law of respective nations and such acts/laws should have content relating to privacy/personal data of smart TV stakeholder/users [53, 54].

Most existing TV program recommendation methods disregard the fact that people often exhibit similar personality traits and tie strengths in terms of different program interests or preferences. In this paper, we proposed an innovative recommendation method called ROPPSA for TV recommendation in social and personality groups. ROPPSA generates similar personalities and tie strengths among TTVs and TV viewers for improving group recommendation accuracy using two different procedures. We carried out an extensive experimentation procedure using a relevant ATU dataset. Experimental results show that our proposed ROPPSA method outperformed other relevant TV program recommendation methods in terms of recommendation accuracy and other utilized metrics.

In future, it is our aim to explore more social-based patterns pertaining to TV program recommendation by applying other social property features involving network centrality measures. Consequently, relevant TV viewers in such a method will be determined by analyzing their information pertaining to network centrality.

Data Availability

The data used in this study are available from the corresponding author upon reasonable request.

Conflicts of Interest

The authors declare no conflicts of interest.

Supplementary Materials

The dataset utilized in this research article involves relevant data collected from students in Accra Technical University (ATU), Ghana, and is an extension from a dataset used in [1]. Data were gathered from students in Higher National Diploma (HND) Marketing (n = 2864), HND Computer Science n = 720), and HND Building Technology (n = 812). Therefore, the ATU dataset utilized in this paper contains a total of 4,396 TV viewers (2,421 males representing 55.07% and 1,975 females representing 44.93%). The ATU dataset in this research article contains tie strength data which comprises computations of contact frequencies (Figure 1) and contact duration (Figure 2) of ATU students (users designated as TV viewers). The ATU dataset also comprises personality trait ratings (scale of 1 to 5) of all users (TV viewers) in accordance with the Big Five Personality Traits (BFPT). The total number of personality ratings for all the traits combined in the ATU dataset is 22,541. Figure 3 illustrates the personality trait data utilized in our ATU dataset for experimentation. Additionally, in order to ensure and substantiate recommendation accuracy in our experimentation procedure, we gathered TV program interest of TV viewers in accordance with the following: movie/TV series-horror, action, comedy, adventure, and drama; sports/games-football, basketball, hockey, boxing, and athletics. Tables 1 and 2 illustrate data regarding TV program interest of TV viewers in the ATU dataset in accordance with their ratings. Figure 1: ATU dataset—contact frequency trends. Figure 2: ATU dataset—contact duration trends. Figure 3: ATU dataset1—personality data. Table 1: ATU dataset—sports TV program ratings. Table 2: ATU dataset—movie/TV series program ratings. (Supplementary Materials)

References

E. Kim, S. Pyo, E. Park, and M. Kim, “An automatic recommendation scheme of TV program contents for (IP)TV personalization,” IEEE Transactions on Broadcasting, vol. 57, no. 3, pp. 674–684, 2011.
View at: Publisher Site | Google Scholar
S. Pyo, E. Kim, and M. Kim, “Automatic and personalized recommendation of TV program contents using sequential pattern mining for smart TV user interaction,” Multimedia Systems, vol. 19, no. 6, pp. 527–542, 2013.
View at: Publisher Site | Google Scholar
S. Pyo, E. Kim, and M. Kim, “LDA-Based unified topic modeling for similar TV user grouping and TV program recommendation,” IEEE Transactions on Cybernetics, vol. 45, no. 8, 2015.
View at: Publisher Site | Google Scholar
M. Bambia, R. Faiz, and M. Boughanem, “Context-awareness and viewer behavior prediction in social-TV recommender systems: survey and challenges,” in Communications in Computer and Information Science, T. Morzy, P. Valduriez, and L. Bellatreche, Eds., vol. 539, pp. 52–59, Springer, Cham, Switzerland, 2015.
View at: Google Scholar
D. Véras, T. Prota, A. Bispo, R. Prudêncio, and C. Ferraz, “A literature review of recommender systems in the television domain,” Expert Systems with Applications, vol. 42, no. 22, pp. 9046–9076, 2015.
View at: Publisher Site | Google Scholar
S. Song, H. Moustafa, and H. Afifi, “Advanced IPTV services personalization through context-aware content recommendation,” IEEE Transactions on Multimedia, vol. 14, no. 6, pp. 1528–1537, 2012.
View at: Publisher Site | Google Scholar
J. Bobadilla, F. Ortega, A. Hernando, and A. Gutiérrez, “Recommender systems survey,” Knowledge-Based Systems, vol. 46, pp. 109–132.
View at: Publisher Site | Google Scholar
Z. Yu and X. Zhou, “TV3P: an adaptive assistant for personalized TV,” IEEE Transactions on Consumer Electronics, vol. 50, no. 1, pp. 393–399, 2004.
View at: Publisher Site | Google Scholar
A. Martinez, J. Arias, A. Vilas, J. Garcia Duque, and M. Lopez Nores, “What’s on TV tonight? An efficient and effective personalized recommender system of TV programs,” IEEE Transactions on Consumer Electronics, vol. 55, no. 1, pp. 286–294, 2009.
View at: Publisher Site | Google Scholar
M. Bjelica, “Unobtrusive relevance feedback for personalized TV program guides,” IEEE Transactions on Consumer Electronics, vol. 57, no. 2, pp. 658–663, 2011.
View at: Publisher Site | Google Scholar
M. Krstic and M. Bjelica, “Context-aware personalized program guide based on neural network,” IEEE Transactions on Consumer Electronics, vol. 58, no. 4, p. 1301, 2012.
View at: Publisher Site | Google Scholar
K. Verbert, N. Manouselis, X. Ochoa et al., “Context-aware recommender systems for learning: a survey and future challenges,” IEEE Transactions on Learning Technologies, vol. 5, no. 4, pp. 318–335, 2012.
View at: Publisher Site | Google Scholar
Y. Hu, Q. Peng, X. Hu, and R. Yang, “Time aware and data sparsity tolerant web service recommendation based on improved collaborative filtering,” IEEE Transactions on Services Computing, vol. 8, no. 5, pp. 782–794, 2015.
View at: Publisher Site | Google Scholar
J. Zhu, J. Zhang, C. Zhang et al., “CHRS: cold start recommendation across multiple heterogeneous information networks,” IEEE Access, vol. 5, pp. 15283–15299, 2017.
View at: Publisher Site | Google Scholar
F. Xia, N. Y. Asabere, H. Liu, Z. Chen, and W. Wang, “Socially aware conference participant recommendation with personality traits,” IEEE Systems Journal, vol. 11, no. 4, pp. 2255–2266, 2017.
View at: Publisher Site | Google Scholar
N. Y. Asabere, A. Acakpovi, and M. B. Michael, “Improving socially-aware recommendation accuracy through personality,” IEEE Transactions on Affective Computing, vol. 9, no. 3, pp. 351–361, 2017.
View at: Publisher Site | Google Scholar
N. Y. Asabere, F. Xia, Q. Meng, F. Li, and H. Liu, “Scholarly paper recommendation based on social awareness and folksonomy,” International Journal of Parallel, Emergent and Distributed Systems, vol. 30, no. 3, pp. 211–232, 2015.
View at: Publisher Site | Google Scholar
F. Xia, N. Y. Asabere, H. Liu, N. Deonauth, and F. Li, “Folksonomy based socially-aware recommendation of scholarly papers for conference participants,” in Proceedings of the 23rd International Conference on World Wide Web-WWW’14 Companion, Seoul, Republic of Korea, April 2014.
View at: Google Scholar
N. Y. Asabere, F. Xia, W. Wang, J. J. P. C. Rodrigues, F. Basso, and J. Ma, “Improving smart conference participation through socially aware recommendation,” IEEE Transactions on Human-Machine Systems, vol. 44, no. 5, pp. 689–700, 2014.
View at: Publisher Site | Google Scholar
F. Xia, N. Y. Asabere, J. J. P. C. Rodrigues, F. Basso, N. Deonauth, and W. Wang, “Socially-aware venue recommendation for conference participants,” in Proceedings of the 110th International Conference on Ubiquitous Intelligence and Computing and 2013 IEEE 10th International Conference on Autonomic and Trusted Computing, pp. 134–141, Vietri sul Mere, Italy, December 2013.
View at: Google Scholar
N. Y. Asabere, B. Xu, A. Acakpovi, and N. Deonauth, “SARVE-2: exploiting social venue recommendation in the context of smart conferences,” IEEE Transactions on Emerging Topics in Computing, 2018.
View at: Publisher Site | Google Scholar
M. Hornick and P. Tamayo, “Extending recommender systems for disjoint user/item sets: the conference recommendation problem,” IEEE Transactions on Knowledge and Data Engineering, vol. 24, no. 8, pp. 1478–1490, 2012.
View at: Publisher Site | Google Scholar
S. Sedhain, S. Sanner, D. Braziunas, L. Xie, and J. Christensen, “Social collaborative filtering for cold-start recommendations,” in Proceedings of the 8th ACM Conference on Recommender systems-RecSys’14, pp. 345–348, New York, NY, USA, October 2014.
View at: Google Scholar
S. Sedhain, A. Menon, S. Sanner, L. Xie, and D. Braziunas, “Low-rank linear cold-start recommendation from social data,” in Proceedings of the 31st AAAI Conference on Artificial Intelligence (AAAI-17), pp. 1502–1508, San Francisco, CA, USA, 2017.
View at: Google Scholar
A. Vinciarelli and G. Mohammadi, “A survey of personality computing,” IEEE Transactions on Affective Computing, vol. 5, no. 3, pp. 273–291, 2014.
View at: Publisher Site | Google Scholar
R. Hu and P. Pu, “Acceptance issues of personality-based recommender systems,” in Proceedings of the 3rd ACM Conference on Recommender Systems-RecSys’09, pp. 221–224, New York, NY, USA, October 2009.
View at: Google Scholar
M. Krstic and M. Bjelica, “Personalized program guide based on one-class classifier,” IEEE Transactions on Consumer Electronics, vol. 62, no. 2, pp. 175–181, 2016.
View at: Publisher Site | Google Scholar
H.-J. Kwon and K.-S. Hong, “Personalized smart TV program recommender based on collaborative filtering and a novel similarity method,” IEEE Transactions on Consumer Electronics, vol. 57, no. 3, pp. 1416–1423, 2011.
View at: Publisher Site | Google Scholar
M. Kompan and M. Bielikova, “Group recommendations: survey and perspectives,” Computing and Informatics, vol. 33, no. 2, pp. 446–476, 2014.
View at: Google Scholar
N. R. Kim, S. Oh, and J. H. Lee, “A television recommender system learning a user’s time-aware watching patterns using quadratic programming,” Applied Sciences, vol. 8, no. 8, pp. 1–14, 2018.
View at: Publisher Site | Google Scholar
M. U. Chaudhry, S. Oh, N. Kim, and J. H. Lee, “Heterogeneous information network based TV program recommendation,” in Proceedings of the 16th International Symposium on Advanced Intelligent Systems, pp. 4–7, Mokpo, Republic of Korea, 2015.
View at: Google Scholar
P. Cesar and D. Geerts, “Past, present, and future of social TV: a categorization,” in Proceedings of the IEEE Consumer Communications and Networking Conference (CCNC), pp. 347–351, Las Vegas, NV, USA, January 2011.
View at: Google Scholar
X. Wang, L. Sun, Z. Wang, and D. Meng, “Group recommendation using external followee for social TV,” in Proceedings of the 2012 IEEE International Conference on Multimedia and Expo, pp. 37–42, Melbourne, Australia, July 2012.
View at: Google Scholar
C. Shin and W. Woo, “Socially aware TV program recommender for multiple viewers,” IEEE Transactions on Consumer Electronics, vol. 55, no. 2, pp. 927–932, 2009.
View at: Publisher Site | Google Scholar
S. Shepstone, Z.-H. Tan, and S. Jensen, “Audio-based age and gender identification to enhance the recommendation of TV content,” IEEE Transactions on Consumer Electronics, vol. 59, no. 3, pp. 721–729, 2013.
View at: Publisher Site | Google Scholar
R. Sotelo, Y. Blanco-Fernandez, M. Lopez-Nores, A. Gil-Solla, and J. Pazos-Arias, “TV program recommendation for groups based on muldimensional TV-anytime classifications,” IEEE Transactions on Consumer Electronics, vol. 55, no. 1, pp. 248–256, 2009.
View at: Publisher Site | Google Scholar
A. Odic, M. Tkalcic, J. F. Tasic, and A. Kosir, “Personality and social context: impact on emotion induction from movies,” in Proceedings of the UMAP 2013, Rome, Italy, June 2013.
View at: Google Scholar
M. Elahi, M. Braunhofer, F. Ricci, and M. Tkalcic, “Personality-based active learning for collaborative filtering recommender systems,” Advances in Artificial Intelligence, Springer, Berlin, Germany, 2013.
View at: Publisher Site | Google Scholar
B. Ferwerda, E. Yang, M. Schedl, and M. Tkalcic, “Personality traits predict music taxonomy preferences,” in Proceedings of the 33rd Annual ACM Conference Extended Abstracts on Human Factors in Computing Systems-CHI EA’15, pp. 2241–2246, Seoul, Republic of Korea, April 2015.
View at: Google Scholar
A. Felfernig, L. Boratto, M. Stettinger, and M. Tkalčič, “Personality, emotions, and group dynamics,” in Group Recommender Systems. SpringerBriefs in Electrical and Computer Engineering, Springer, Cham, Switzerland, 2018.
View at: Google Scholar
M. Tkalčič, B. D. Carolis, M. de Gemmis, A. Odić, and A. Košir, “Introduction to emotions and personality in personalized systems,” in Emotions and Personality in Personalized Services. Human–Computer Interaction Series, M. Tkalčič, B. De Carolis, M. de Gemmis et al., Eds., Springer, Cham, Switzerland, 2016.
View at: Google Scholar
M. Tkalčič, “Emotions and personality in recommender systems,” in Proceedings of the 12th ACM Conference on Recommender Systems, pp. 535-536, New York, NY, USA, 2018.
View at: Google Scholar
L. Quijano-Sanchez, J. A. Recio-Garcia, and B. Diaz-Agudo, “Personality and social trust in group recommendations,” in Proceedings of the 22nd IEEE International Conference on Tools with Artificial Intelligence, pp. 121–126, Arras, France, 2010.
View at: Google Scholar
L. Quijano-Sanchez, J. A. Recio-Garcia, and B. Diaz-Agudo, “Using personality to create alliances in group recommender systems,” in Proceedings of the International Conference on Case-Based Reasoning, pp. 226–240, Springer, London, UK, September 2011.
View at: Google Scholar
L. Quijano-Sanchez, J. A. Recio-Garcia, and B. Diaz-Agudo, “Happymovie: a Facebook application for recommending movies to groups,” in Proceedings of the IEEE 23rd International Conference on Tools with Artificial Intelligence, pp. 239–244, Boca Raton, FL, USA, 2011.
View at: Google Scholar
L. Quijano-Sanchez, J. A. Recio-Garcia, and B. Diaz-Agudo, “An architecture and functional description to integrate social behaviour knowledge into group recommender systems,” Applied Intelligence, vol. 40, no. 4, pp. 732–748, 2014.
View at: Publisher Site | Google Scholar
J. A. Recio-García, L. Quijano, and B. Díaz-Agudo, “Including social factors in an argumentative model for group decision support systems,” Decision Support Systems, vol. 56, pp. 48–55, 2013.
View at: Publisher Site | Google Scholar
J. A. Recio-Garcia, G. Jimenez-Diaz, A. A. Sanchez-Ruiz, and B. Diaz-Agudo, “Personality aware recommendations to groups,” in Proceedings of the Third ACM Conference on Recommender Systems-RecSys’09, pp. 325–328, New York, NY, USA, October 2009.
View at: Google Scholar
R. Prada, S. Ma, and M. A. Nunes, “Personality in social group dynamics,” in Proceedings of the 12th IEEE International Conference on Computational Science and Engineering, pp. 607–612, Vancouver, Canada, August 2009.
View at: Google Scholar
J. Wu, Y. Shi, and C. Guo, “A resource recommendation method based on user taste diffusion model in folksonomies,” International Journal of Knowledge and Systems Science, vol. 3, no. 1, pp. 18–30, 2012.
View at: Publisher Site | Google Scholar
J. B. Schafer, D. Frankowski, J. Herlocker, and S. Sen, “Collaborative filtering recommender systems,” Lecture Notes in Computer Science, vol. 4231, Springer, Berlin, Germany, 2007.
View at: Google Scholar
E. Baykan, M. Henzinger, L. Marian, and I. Weber, “A comprehensive study of features and algorithms for URL-based topic classification,” ACM Transactions on the Web, vol. 5, no. 3, pp. 1–29, 2011.
View at: Publisher Site | Google Scholar
K. Irion and N. Helberger, “Smart TV and the online media sector: user privacy in view of changing market realities,” Telecommunications Policy, vol. 41, no. 3, pp. 170–184, 2017.
View at: Publisher Site | Google Scholar
M. Ghiglieri, M. Volkamer, and K. Renaud, “Exploring consumers’ attitudes of smart TV related privacy risks,” in Proceedings of the 5th International Conference on Human Aspects of Information Security, Privacy, and Trust, pp. 656–674, Vancouver, Canada, July 2017.
View at: Google Scholar

Copyright

Copyright © 2020 Nana Yaw Asabere and Amevi Acakpovi. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

PDF Download Citation

Download other formats

Order printed copies

Views

663

Downloads

717

Citations