| With the proposal of the concept of Internet plus, its application has been widely used on different fields in our country. People’s requirements for knowledge are more and more strong. However, massive information appear on the Internet, and it is more and more difficult for people to obtain the effective information. Therefore, it is still an urgent issue how to quickly and accurately retrieve useful information from mass data. To solve this problem, many scholars use the text compression technique to extract useful information, and it is also called text summarization.In this paper, we deeply study the current related research about text summarization technologies in the world. Furthermore, we consider the influence of the semantic factors on the text summarization and propose a novel summarization technology based on conceptual object model. In this paper, the main research content is listed as follows:(1) Firstly, we disambiguate the ambiguous words in the documents, and then merge the synonym into a series of synonym phrases and extract a number of concepts to describe them. Secondly, the concepts are used to build up the vector space model of the sentences in the documents.(2) In this paper, we use the improved k- means algorithm to cluster the sentences. According to the similarity, the sentences are divided into several groups, the sentences similarity in the same group is high, the similarity in different groups is low. Finally, we construct a conceptual object model based on the clustered results.(3) This paper construct two graph models, one is based on the concept-based sentences, the other is based on conceptual object model. In the process of the graph model construction, the impacts of syntactic features and the sentences with the same semantics on graph edges are carefully considered. Furthermore, we build up another graph model by regarding the conceptual object model as the graph node. In the process of calculating the weight of the sentence, we modify sentence score which is based on 1) the weight of conceptual object model and 2) the importance of sentence position.The innovations of this paper are:(1) We use the improved k- means algorithm to cluster the sentences, and then select the sentences with higher weights to initialize cluster center. Motivated by the idea of the object oriented concept, conceptual object model is proposed in this paper. We regard these clustered results as a number of conceptual objects, the semantic similarity of sentences in the same object is high, and the semantic similarity in different concepts is relatively low. Finally, we extract the related association attributes and behaviors from the sentences to construct a conceptual object model.(2) Aiming to the shortcoming of sentence-based graph model, this paper, considers the impacts of syntactic features and the sentences with the same semantics, and uses the concept-based sentences and conceptual object model to construct a graph model, then use the constructed graph model to deal with the text summary. Experimental results show that the quality of the generated summarization is significantly improved.Finally, we create a new corpus by extending the multi-document corpus of HIT-CIR. we compare the proposed text summarization technologies based on conceptual object model with the method based on sentence-based graph model on the multi-documents set. Experimental results demonstrate that the proposed method performs better than other methods on both ROUGE-1 and ROUGE-2. That is to say, text summarization technologies based on conceptual object model can effectively improve the quality of the text summarization. |
