Writing Assistant System Based On Topic Recommendation

With the improvement of internationalization, English learning is becoming more and more important. In the process of language learning, writing often plays a very important role. Writing can highlight the mastery of language proficiency, if you want to write high quality articles, you must have a lot of reading experience and knowledge base. However, for non-native English students, due to the lack of English reading experience and knowledge base, writing high quality English articles is not an easy thing. In recent years, many different writing assistant systems have appeared, which provide a great help for the users. But most systems focus on lexical, phrasal and sentential level, and there is no effective help at the theme level. They don’t provide tips and recommendations about the topic of content and can’t effectively solve the problem that the users lack of material or inspiration. So in this paper, we analysis the advantages and disadvantages of existing writing assistant systems and combine with present users’ demand. In the end, we design and implement an English writing assistant systems which is based on the topic of content. Specifically, this paper mainly consists of the following three aspects.The first aspect is that in order to analyze the topic of the article simply, directly and effectively, this paper studies topic word extraction. Firstly, this paper introduces the traditional topic word extraction methods. Secondly, using the topic model LDA that is combined with the traditional methods, we put forward the topic word extraction method based on the fusion model of LDA. Finally, through the experiment, we verify that to some extent the topic information of word which is mined by the topic model LDA can improve the effect of topic word extraction. We use the above methods to extract topic words that are mainly used for retrieving the related articles and computing the similarity of the articles in the writing assistant system based on topic recommendation.The second aspect is that for the purpose of computing sentence similarity, this paper firstly analyzes the advantages and disadvantages of several methods. Then, we research and implement the sentence similarity computation based on word embedding. Experiments show that word embedding contains rich information of words such as a latent feature of the word, syntactic and semantic properties which make the method get a better result. In the implement of system, we apply the above methods about sentence similarity computation to compute the similarity between sentences in article and the input from users so that the system recommends sentences to them.The last aspect is that in this paper, we design and implement an English writing assistant systems which is based on the topic of content, evaluate the performance of the system and test the response time of the system. The system is composed of three different modules: word collocation module, real-time hit of example sentence module and topic-based sentence recommendation module. It can help users at the theme level and to some extent solve the problem that the users lack of material or inspiration. In order to prove the practicality of the model, we evaluate the result of the system’s modules in the manual or mechanical way. Finally, we test the response time of the system and the test result indicates that the real-time property of the system can be satisfied.