| Input: |
| (i) The candidate topic discriminative terms of a given document. |
| Output: |
| (ii) The set of topical clusters with topical describing information. |
| Procedure: |
| (1) Computing the re-occurrences topic span for each candidate topic discriminative term. |
| (2) Ordering by the length of re-occurrences topic span for each candidate topic discriminative term. |
| (3) repeat |
| (4) Proceeding iteratively, starting with the longest re-occurrences topic span of TDT, ending with |
| the shortest and last TDT. |
| (5) Judging the overlap relationship of arbitrary two TDT’s [, ]. |
| (6) if (Two topic span intervals of TDT are intersecting) |
| (7) Two vertices of corresponding topic span intervals of TDT will be directly connected. |
| (8) else |
| (9) Two vertices will respectively belong to different topic span intervals, namely belong to |
| different subgraphs. |
| (10) until all independent subgraphs are constructed; |
| (11) Labeling each subgraph for , where . |
| (12) All independent subgraphs will be connected with the vertices that are directly proximal |
| in the corresponding TDT’s topic span, so a whole topical graph is constructed. |
| (13) repeat |
| (14) if (there exists one TDT which is monosemous) |
| (15) Select the higher weight one as initial vertex. |
| (16) Iteratively calculate the similarity of topic chains between this vertex and other ones with its |
| neighbors by formula (3). |
| (17) else |
| (18) Iteratively calculate the similarity of topic chains from the maximal weight of two |
| vertices by formula (4). |
| (19) until all candidate TDTs only exist the unique topic chain; |
| (20) Updating the weight of edges by formula (5) and pruning completely irrelevant edges in |
| the topical graph. |
| (21) Readjusting a topical conflict interval between two vertices of corresponding TDTs which |
| are connected by the pruned edge. |
| (22) S <- choosing the vertices of the maximum degree in the topical graph. |
| (23) Integrating the other vertices into topical clusters according to the adjacency and |
| similarity relationship. |
| (24) Ignoring the isolated individuals and too small topical clusters. |
| (25) Generating topical describing information to implement topic identification. |
| (26) Return The set of topical clusters with topical describing information. |
