An Intention-Topic Model Based on Verbs Clustering and Short Texts Topic Mining

Microblog, Twitter, status messages, the classified information website and so on are experiencing explosive growth with the development of web2.0, people prefer to use short texts to express their intentions and activities. Yet, when people submit some requirements through short texts, they hope to get a feedback which can help them to solve their problems rather than relevant content. Sometimes people need corresponding intention rather than similar content. However, current researches cannot solve the problem well. In this paper, we propose an intentiontopic model: Verb-Biterm Topic Model(V-BTM), which aims at corresponding intention matching. Intention is expressed by verbs and topic is expressed by BTM. Intention is the action of people want to express and topic is the goal of the intention. The key of the model is that people tend to express their intention with verbs and tend to express the topic with non-verb. In this model, firstly, we distinguish intentions with the verb clustering with the help of word2vec which is a deep learning tool. Secondly, we mine the topic using Biterm Topic Model(BTM) on the data without verbs. We carry out experiments on real-world short text collections. The results demonstrate that our approach can get better verb clustering and mine more coherent topics. Furthermore, the new model can be the base of our future researches.