Research On Expert Recommendation Mechanism And Answer Summarization Algorithm For CQA

Community Question Answering(CQA)provides an important interactive platform to share and acquire knowledge for users,which they could post and answer questions.With the increasing number of users in CQA,askers may take several days to get the answer while responders are not interested in questions assigned.To solve this problem,scholars have proposed a variety of expert recommendation mechanisms.However,most of them suffer from delayed answers and low best answer coverage.In addition,a non-factoid question in CQA often contains several answers,but the best answer is not complete.As deep learning techniques have been widely used in text summaries,we attempt to introduce this to CQA for answer summaries.However,the current text summarization algorithms still have some problems such as semantic incompatibility and self-repetition.In view of the lack of existing research work,the specific work done in this paper is as follows:(1)We propose a graph-based topic sensitive answerer rank algorithm(TSAR)with DSSM.DSSM is mainly used to have a semantic analysis between question text and user text while TSAR could be applied to conduct a link analysis on users' interaction feature.We utilize the combination of DSSM and TSAR to find an appropriate expert for the newly posted question in CQA.First,we calculate the similarity of texts by DSSM,which generates a ranked list of candidates.Based on this,we build a weighted directed graph.Second,we use our proposed TSAR to estimate the authority score of each node in the graph so as to recommend a final list of experts.(2)We propose an answer summary algorithm based on the multi-layer attention mechanism(ASMAM).In order to further improve the ability of text representation,the self-attention mechanism and the multi-head attention mechanism are introduced in sentence encoding and text encoding respectively.Gated Recurrent Unit(GRU)is used in both the encoder and decoder,which could solve "Gradient Disappearance" of Recurrent Neural Networks(RNN)and too many parameters of Long Short-Term Memory(LSTM).In order to avoid self-repetition due to the hidden state of the decoder,we introduce the intra attention mechanism.(3)Finally,we apply the algorithm proposed in this paper to the prototype system.When a user submits a new question to the CQA,the expert recommendation mechanism is used to generate expert candidates who are capable,interested of it and can respond in a timely manner.For several answer texts corresponding to the non-factoid question,the answer summary module in the system is used to summarize and condense answers,so that users can quickly get a comprehensive answer.