Modeling Topical Relevance for Multi-Turn Dialogue Generation

Topic drift is a common phenomenon in multi-turn dialogue. Therefore, an ideal dialogue generation models should be able to capture the topic information of each context, detect the relevant context, and produce appropriate responses accordingly. However, existing models usually use word or sentence level similarities to detect the relevant contexts, which fail to well capture the topical level relevance. In this paper, we propose a new model, named STAR-BTM, to tackle this problem. Firstly, the Biterm Topic Model is pre-trained on the whole training dataset. Then, the topic level attention weights are computed based on the topic representation of each context. Finally, the attention weights and the topic distribution are utilized in the decoding process to generate the corresponding responses. Experimental results on both Chinese customer services data and English Ubuntu dialogue data show that STAR-BTM significantly outperforms several state-of-the-art methods, in terms of both metric-based and human evaluations.

Research and Implementation of Basic Education Resource Retrieval Strategy Based on the Relationship among Knowledge Points

In order to solve the lack of knowledge relevance and topic drift in traditional keyword search, this paper proposed knowledge extraction strategy based knowledge point association and recognition of user intention based user intention theme map and the mapping from resource to knowledge points strategy .It realized the knowledge sestematic and relevant of the search results and improved the retrieval accuracy.

TOPIC TRACKING WITH IMPROVED REPRESENTATION MODEL AND JOINT TRACKING METHOD

Topic tracking is to monitor a stream of stories to find additional stories on a topic identified by several samples. However, the predefined information about a tracked topic does not provide enough information to deal with the new information occurred in the tracking procedure. To overcome this problem, we proposed a joint tracking method using both the topic-specific information from the predefined information and the non-topic-specific information from the data on other topics. Besides, to overcome the limitation of the representation model and the topic drift problem, we have also used two other improvements: a topic-based weighting method is used to measure the features of both tracked topics and single testing stories; a dynamic topic model is extended with the information brought by the incoming related stories and the noise is filtered out with the information in the incoming unrelated stories. The implemented tracking systems are evaluated on the Chinese subset of TDT4 corpus by the TDT2003 evaluation method. The experimental results indicate that the above methods all improve the tracking performance. More importantly, these techniques are complementary to one another and not mutually exclusive.