The Research Of Deep Joint Models For Answer Selection And Question Recommendation On CQA Platforms

Community Question Answering(CQA)sites have emerged as a type of popular social platforms.In the social network,users can raise questions to leverage the expertise of other users participating in the community.Compared with serch engines,CQA sites can fulfill users' demands in an easier way.Given the large amount of knowledge organized in the format of question-answers,users suffer from information overload.A research problem is to automatically select correct answers for a given question so as to improve intelligence for CQA and minimize the time for satisfying users seeking the correct answers.Another interesting research problem is to recommend questions to users so as to maximize their engagements with the platform.There are some challenges for the two tasks.First,the two both encounter the data sparsity problem.For answer selection task,the data sparsity problem comes from that the words in the question and answer sentences are not match well even though they are semantically related.For question recommendation task,the number of questions interacted with a user is very limited compared with the number of the whole question set.Second,there are heterogeneous information sources on CQA platforms.It's not trivial to make full use of those information sources to alleviate the data sparsity problem.In this thesis,we propose joint models that incorporate different information sources with deep learning methods for the two tasks on CQA.1.For answer selection task,we propose a hybrid attention mechanism to model the question-answer pairs,we also model users from their written answers.Specifically,for each word in question/answer,we calculate the intra-sentence attention indicating its local importance and the inter-sentence attention implying its importance to the counterpart sentence.The inter-sentence attention is based on the interactions between question-answer pairs,and the combination of these two attention mechanisms enables us to align the most informative parts in questionanswer pairs for sentence matching.Additionally,we exploit user information for answer selection due to the fact that users are more likely to provide correct answers in their areas of expertise.We model users from their written answers to alleviate data sparsity problem,and then learn user representations according to the informative parts in sentences that are useful for question-answer matching task.2.We propose a model that is able to jointly model both implicit and explicit information for question recommendation.The model integrates multiple data sources and addresses the problem of data heterogeneity.In the proposed model,we dynamically discover latent user groups and incorporate those hierarchical information to bridge the semantic gaps among users in the shared latent space.We evaluate the proposed model on two real-world datasets,and demonstrate that our model outperforms state-of-the-art alternatives by a large margin.We also investigate different structures of the proposed model to study the effects of different data sources.3.We propose to address question recommendation task that involves heterogeneous information spaces,namely interactive space(i.e.user-item),structural space(user-user)and semantic space(user-attribute).Instead of modeling each information space independently,we propose to seamlessly integrate information across heterogeneous spaces.To do this,we propose an attention mechanism where the users attend differently to their social neighbors(structural space)for learning comprehensive user representations.The attentions are parameterized by the interactions between user profiles(semantic space),and they are collaboratively learned for the recommendation task(interactive space).In this way,information across different spaces can be complementary to each other for boosting recommendation performance.We also prove that the proposed attentive user representation learning method is a generalization of traditional social regularization and network embedding methods.We validate the effectiveness of the proposed model with two real datasets,and show that the proposed model is able to outperform state-of-the-art recommender models significantly.