Research And Implementation Of Enterprise Recommendation Technology Based On Deep Learning

Recommendation system is a kind of information that can provide users with information or items that meet their own needs or are liked by users according to their historical information or their own characteristics from mass information.Based on the historical business records of enterprises or the characteristics of enterprises themselves,it can screen the useful information that enterprises can use for business from the mass of information,so as to save time and labor costs and promote the transformation of traditional enterprises to informatization.Based on this practical requirement,this paper designs and implements an enterprise recommendation system based on deep learning,which is composed of distributed data acquisition system,enterprise classification algorithm based on deep neural network and enterprise recommendation algorithm based on content.In terms of distributed data collection,this paper adopts the Scrapy framework under Python.According to the characteristics of Redis memory database,Mysql database is used as the persistent storage of data to complete the scrapy-redis distributed crawler system.For the enterprise classification algorithm based on deep learning,because the original data captured from the network does not contain the category information of the enterprise,the attribute of the enterprise's category information is crucial in the recommendation.In this part of the thesis,three kinds of neural networks,feedforward neural network,convolutional neural network and n-gram neural network are completed under the condition of limited data sets.The accuracy of the three neural networks in the verification set is 86.28%,86.16% and 86.16% respectively.Feedforward neural network is used to realize enterprise classification.Aiming at the recommendation algorithm of enterprises based on content,this thesis formulates the calculation method of enterprise similarity based on the actual business of enterprises.The description methods of enterprise attributes are specified respectively,and the calculation methods of three different fields are defined.That is,the list class field,the text class field,and the number type field.For the most important text class fields,this thesis implemented LSI model and Word Embedding method,and compared the final calculation structure with Baidu AI.The pairwise Pearson similarity of LSI model,Word Embedding model and Baidu AI was 0.3979,0.1984 and 0.6451.Finally,according to the enterprise business selection LSI model text class field classification.In the final recommendation results,based on the feedback of the final enterprise,the success rate of the business conducted according to the recommendation system and the business conducted without the recommendation system is about 7.5% higher than that of the business.