A Study On Web Page Recommendation Algorithm Based On Named Entity

With the rapid development of the network technology and computer technology, online information increases exponentially in bandwidth. In such an explosive information society, people want to get much information through the network, and in this background search engine was born and developed. Generally speaking, a customer inquires information through the search engine. Search engine can offer relevant sorting web according to a certain algorithm, although sometimes it can not meet user queries. In order to provide satisfactory service for customers, this paper mainly studies a new intelligent recommendation item for customer which is based on user queries behavior mined from inquire logs. A study on web page recommendation algorithm based on named entity is put forward, and it can offer better recommendation pages.A large number of queries contain more or less entities, which are divided into traditional named entity and special named entity in inquire logs. The traditional named entity includes personal names, location names and organization names. And special named entity which has a special meaning is closely related to human life, such as TV, environment, economic, movie, medicine, transportation, IT and education fields. For the entity recognition and type identification of inquire logs and documents, we use different methods, especially identifying entiy type of inquire logs by using triad probability method. The entity according to the type of entity will be mapped to page type to get better web recommendation. This paper puts forward web navigation links based on hybrid markov model for catalogue pages and topic recommendation algorithm based on LDA feature selection for theme pages. According to the current user’s clicking behavior, the item will be recommended less than eight inter-navigation links for the customers; moreover these links reflect the most customers’ information demand and contain the same entity according to the query demands, thus it can reduce searching time and bandwidth consumption for users. A document is composed of several topics, therefore this paper comes up with a web recommendation algorithm based on LDA feature extraction. We can model texts by using LDA model, obtain sub-LDA models of each category, and assign text categories. Finally, we recommend the web links to users in terms of entity type and the matching degree of queries.According to the experiment’s results, a recommendation algorithm based on hybrid Markov link model achieves a comparative satisfactory effect. Moreover, recommendation of LDA feature selection is superior to conventional recommendation algorithm and meets the needs of the users for searching information.