Movie Review Summarization Using Supervised Learning and Graph-Based Ranking Algorithm

Table 4

Movie review classification accuracy on three tasks.

	Features	PL04	Full IMDB	Subjectivity
1	Unigrams with NB	81.5	86.66	90.75
2	Bigrams with NB	77.7	88.29	76.03
3	Unigrams + bigrams with NB	82.4	88.91	91.22
4	Unigram frequency + smoothed IDF + cosine normalization	82.1	87.36	90.7
5	Bigram frequency + smoothed IDF + cosine normalization	81.15	88.31	76.72
6	Unigrams + bigrams + smoothed IDF + cosine normalization	83.7	89.28	90.91
10	Benchmark model [62]	88.90	88.89	88.13

PL04 refers to the collection of 2000 movie reviews often used as benchmark dataset for sentiment classification [61], Full IMDB dataset is a collection of 50,000 reviews, and sentence subjectivity dataset is a collection of 1000 movie reviews [61].