Research on Answer Ranking in Online Academic Communities Based on Multi-Label Classification Technology from the Perspective of Text Topics
[Purpose/Significance]The uneven quality of user-generated answers in online academic commu-nities makes it difficult for users to obtain efficient decision support.Filtering high-availability answers can promote the efficient use of question and answer knowledge in online academic communities.[Method/Process]From the perspective of text topic semantics,this paper proposed a question and answer correlation calculation based on a deep pre-training language model and multi-label classification technology,which was used to achieve the useful-ness ranking of user generated answers.It first extracted the semantic vectors of question and answer text,and then further mapped them into a field-specific topic vector space,thereby realizing the calculation of topic similarity between questions and answers.[Result/Conclusion]Taking all the questions and answers under the"Help Com-pletion"of the thesis submission in"Xiaomuchong"academic community as experimental data,it uses NDCG and Q-Measure as evaluation indicators,and compares with two conventional semantic-based sorting methods such as Cross-Encoder and Bi-Encoder.Experiment result shows that the performance of the proposed method is equivalent to that of conventional methods,but requires less annotation data.
NETL
NSTL
万方数据
