Reasearch On Service Recommendation Algorithm Based On Dark Data

The hottest vocabulary in today's society is big data.It is not only studied by academia,but also sought after by the government.It is also highly valued by the business community.The whole society has entered the era of "big data",which seems to be the strongest driving force to promote social development.By introducing the dark data with high proportion and great value but difficult to collect,analyze and apply,this paper focuses on the ways and methods of recommending the application of dark data.At the same time,there is explosive growth of information on the Internet every day.The personalized service recommendation system can make users get the desired information quickly and conveniently.In this paper,the user mobile photo album is used as the dark data set.Firstly,unstructured short texts are constructed into triples by entity relationship extraction,and then similar or identical nodes are fused by knowledge fusion to establish knowledge map.Then,the keywords obtained by the improved TextRank algorithm are expressed as word vectors in the subject model after Word2 Vec training,and the candidate entity sets are compared with those obtained from WikiPedia corpus to get the final candidate entity and complete entity disambiguation.Through the integration with the external knowledge base,the local personalized knowledge base is obtained.And the PTransE algorithm is used to construct the triple of the relationship path,embedding entities and relationships into low-dimensional space.Through AP algorithm,the relationship between the edges of the existing knowledge map is clustered to complete and predict the knowledge map.Based on the above research results,a service recommendation prototype system based on dark data is designed and developed to realize personalized service recommendation.Because most of the data in the Internet exist in unstructured form,there is no uniform standard to express,so it is very difficult to obtain and clean up the data.In the stage of building knowledge map,we define and maintain a user dictionary to realize network words that can not be effectively recognized by natural language processing tools,so as to extract entity relations more conveniently.In the phase of entity disambiguation,we innovatively use the improved TextRank algorithm with TF-IDF as the weighting factor.For link prediction,we use translation-based representation learning to construct a triple for the relationship path of knowledge maps and embedding it in low-dimensional space.Finally,the feasibility of the method is effectively proved by a series of experiments,and the significance of this topic is proved by the design of service recommendation system.