Research On Potential Customers Based On Company Announcement Knowledge Graph

Company announcement is the official document issued by the company’s daily operation,and is an important reference for industry development and target decision-making.With the increasing number of company announcements,many companies are exploring the direction to find the internal important information through artificial intelligence.Mining and utilizing the knowledge of company announcement is a very important and meaningful task.At present,there are still the following challenges:(1)the amount of company announcement data is huge,and it will continue to increase with the development of time;(2)there are many sources of announcement data,the authenticity is uncertain,and the data types are complex.(3)There is a lot of business value in company announcement.How to find potential customers through enterprise knowledge is also an important research direction.Through the above analysis of the problem,the following three aspects will be targeted to solve the problem:1)Research on named entity recognition for company announcements.The standard corpus is formed by manual annotation,with a total of 1579506 samples,11 entity types and 2197738 entities;the word embedding model is constructed by word2vec,Elmo and Bert methods,and as input,the named entity recognition task is performed by BiLSTM+CRF model.After the experiment,it is found that the best effect of BERT+BiLSTM+CRF model is 83.05%,and the F1 value is 83.36%;the worst model is based on word2vec word embedding model,and the accuracy rate is 80.93%,and the F1 value is 80.87%.2)Research on entity relationship extraction for company announcement.Through the manual way to label and audit the entity relationship of company announcement data,and form a standard corpus,a total of 623019 relationship data.The accuracy of pipeline method based on BERT+BiLSTM+CRF is 73.56%and the F1 value is 73.20%.The method of PCNN+ATT is used to extract direct relationship,and the experimental results show that the accuracy is 74.28%and 73.77%respectively.3)Using neo4j graphic database for company announcement refers to the construction and analysis of atlas database.Finally,a total of 126690 purchasers are counted through network nodes.276 and 117 potential customers are mined by using point centrality algorithm and eigenvector centrality algorithm,and the number of transactions,regional distribution,company type and company industry of customers are visualized.Thus,through the construction of enterprise knowledge graph,we can find potential customers for the company,which is convenient for customer maintenance and new customer development in the future.