Research On Scheme Recommendation Method Based On Improved Collaborative Filtering

In recent years,with the concept of Internet + and Industry 4.0,the Internet technology and scale have been greatly improved.And massive information has appeared in people's field of vision.The era of information explosion makes it difficult for people to obtain the information they need while accepting multiple aspects of information.In response to this situation,traditional search engine technology can help people filter and filter out irrelevant information,but some keywords still do not meet some of the user's needs.So the search results will be disappointing.In response to this problem,personalized recommendation methods has been improved.And the field of e-commerce,such as Jindong,Ali,headline news,etc.have been widely used the recommendation methods.However,there are few recommended technologies in the manufacturing and service industries.At present,the main manufacturing solutions have been applied in the area of the manufacturing industry,the social-driven learning program,and the service recommendation scheme for crowdsourcing distribution.But the technology adopted is the service matching recommendation technology.The mature collaborative filtering technology has not been applied in real terms.If the enterprises adopt strong recommendation technology,it can help enterprises to explore new users and retain old customers,thus improving the company's revenue.Collaborative filtering technology is one of the most mature technologies in the current application.Its core is the collaborative filtering algorithm.It filters the nearest neighbor set according to the user's rating value and similarity algorithm for the product or service,and selects the user with score of high similarity.Recommendations are given to products or services that are not rated by the target user.In the implementation process,the traditional collaborative filtering algorithm is often accompanied by data sparseness and cold start problems,as well as the impact of the user feature tag and the large amount of user scoring data on the recommendation results and efficiency.Aiming at these problems,this paper used the similarity calculation and prediction formula to calculate the user similarity and predict the vacancy data in order to reduce the data sparsity.Based on the user interest feature calculation,the new user's neighbor set was obtained and the appropriate scheme recommendation was given to solve the user cold start problem.It is difficult to give reasonable recommendations based on massive user data.An attribute reduction method using multi-granular rough sets was proposed to streamline data and solve data redundancy problems.The research methods of this paper were described in detail below.(1)In order to solve redundancy problem caused by data,the method of traditional rough set attribute reduction can't give accurate reduction.The multi-granular rough set was adopted to reduce data and reduce data dimension.(2)Based on the data sparsity caused by users' unrated for programs,the Pearson similarity,cloud similarity and the comprehensive similarity measure of the former two methods were adopted.The relevance of the program score was calculated to predict the vacancy data.The average absolute error value of the set was calculated and the similarity method with the lowest error value was selected to predict and to fill the sparse scheme score data.(3)Considering the differences of user interest preferences and subjectivity,the fuzzy clustering algorithm was used to cluster the user feature to form the user cluster.Then the center of user cluster was selected as the representative user to calculate the interest similarity of the new user.And the programs of existing users with high similarity were recommended to new users.The cold start problem was solved to improve the efficiency of the program recommendation.In the case part,take the recommendation of an enterprise car scheme as an example.According to the user data and other scoring data sets,the recommendation method was proposed in this paper.And compared with other recommendation algorithms,it was found that the prediction results were more accurate when adopting the selected methods to predict and fill data.Moreover,when the number of programs increased during the recommendation process,the recommended results were more accurate.Moreover,based on the same number of data sets,different recommended methods were used to give the recommendation results.It was found that the algorithm execution time was also much better than other algorithms.And the method has significantly improved the accuracy and efficiency of the recommendation.In addition,the recommended technology used in this paper could help enterprises to explore potential users,convert them into actual users,improve the efficiency of users' query information and user retention rate and help enterprises to target users and improve enterprise efficiency...