Commodity refined recommendation method based on comment integration mining

Abstract

According to the commodity recommendation method based on comment mining, comment mining and a traditional collaborative filtering recommendation method are combined, users and commodities are analyzed from the aspects of user preferences and commodity characteristics, and the problems of data sparsity and recommendation accuracy are relieved. The method specifically comprises the steps that firstly, user preferences are obtained by mining and analyzing user comments, then the similarity between users is calculated according to the obtained user preferences, and the adverse effect of data sparsity on calculation of the similarity between the users is relieved; secondly, by mining and analyzing commodity comments, a characteristic model of the commodities is constructed, when the commodity similarity is calculated, based on the commodity characteristics, the accuracy of commodity similarity calculation is improved by improving a method for calculating the similarity between the commodities, and the recommendation effect is improved; and thirdly, collaborative filtering based on comment mining and users and collaborative filtering based on comment mining and commodities are combined to generate a mixed model for recommendation, so that the efficiency and the accuracy are greatly improved.

Description

technical field [0001] The invention relates to a refined commodity recommendation method, in particular to a refined commodity recommendation method based on review integration mining, and belongs to the technical field of intelligent commodity recommendation. Background technique [0002] With the rapid development of the mobile Internet, human beings have entered the information age, and e-commerce has been integrated into people's daily life. Online shopping has become the main shopping method for more and more consumers. E-commerce has the advantages of being convenient, fast, cheap and affordable, and not subject to geographical restrictions. Due to the huge number of products, it is difficult for users to choose the products they really need from the huge number of products. At present, more and more online shopping platforms are beginning to emphasize user participation. They encourage users to express their opinions on purchased products. Users convey their opinion...

AI Technical Summary

Problems solved by technology

[0009] First, it is difficult for e-commerce platforms to understand the content that users are interested in. Faced with the huge number of products on e-commerce platforms, it is difficult for users to choose the products they really need from the huge number of products. Most netizens will search for products that have reviews before buying them. , review information will have a significant impact on their purchase behavior, but in a limited time, faced with massive and unstructured review content, it is difficult for users to identify the most valuable information for themselves

There is an urgent need for a convenient product recommendation system that can analyze customer preferences and meet their needs for products. For e-commerce companies, it is necessary to overcome the adverse effects of information overload and recommend products to users when they browse the platform. The right merchandise becomes a serious challenge;

[0010] Second, with the rapid development of e-commerce, the types and numbers of commodities in shopping websites are increasing continuously. It is difficult for users to find the commodities they are interested in in a short time. Commodity recommendation provides an effective solution. The existing collaborative filtering algorithm is Recommendation systems are affected by data sparsity and cold start problems

Both user-based collaborative filtering and product-based collaborative filtering involve the calculation of the similarity between users or products. These similarity calculation methods are calculated based on the scoring items intersected between users. , the common ratings between users will be very small, or even no common ratings, which will greatly affect the similarity calculation results. In traditional collaborative filtering algorithms, when calculating the similarity of users or products, only the rating item sets between users are used. The calculation has great limitations. The sparseness of the user product rating matrix leads to inaccurate or even impossible calculation of the similarity between users or products, which affects the recommendation accuracy of the collaborative filtering algorithm;

[0011] Third, the current search engines are able to search for the relevant information that users need, but their ability to integrate content is poor. They can search for relevant content that users want, but they cannot further sort out the key information that users need. For e-commerce The same is true for the product review information on the platform. The existing technology cannot combine the review mining technology to mine the product characteristics and personal preferences that users care about in the reviews, and cannot meet the needs of users and accurately recommend suitable products for users; Comments emerge. These comments contain a lot of subjective emotions of users. By browsing these comments, we can get a general idea of ​​the public's views on a certain commodity. However, such comment information is rapidly expanding, with a large number and no fixed text structure. It is almost impossible to collect and process massive amounts of information manually, and there is an urgent need for a method that can help users quickly obtain the focus of attention;

[0012] Fourth, in terms of product feature extraction, although manual definition is highly accurate, it requires the participation of domain experts, and different domains require different domain experts. When domain terms are updated, the manual definition method is inefficient

The accuracy of feature extraction in the Chinese field is not high, and there are great difficulties in practical applications; in terms of emotional word extraction, it is too simple to extract adjectives from a string of character strings before and after the feature word as user evaluation emotional words. Considering the corresponding relationship between idiosyncratic words and emotional words, there will be more extraction errors

The method of extracting trait-emotional word pairs through the method of generating language patterns through supervised sequence patterns is only suitable for situations where the sentence structure is simple and the distance between trait words and emotional words is close, and only the part of speech of emotional words is limited to adjectives, still There are certain limitations;

Images