Research And Optimization For Content-based Association Online Commodity Recommendation System

With the rapid development of Internet,information gradually increased,the amount of data grows exponentially,resulting in the announcement of big data era's arrival.Under such environment,it comes up with the information overload problem,the difficulty of obtaining effective information becomes more prominent for its sharp promotion.The way to solve this problem has become a hotspot of now day's research gradually,the recommendation system is a powerful measure to solve the problem,the recommender system can introduce items that certain user will be interested according to the project information or user information,through the matching and similarity calculation.In the research the current recommendation system,there are three kinds of recommendation were raised up: content-based recommendation algorithm,collaborative filtering-based recommendation algorithm and hybrid recommendation algorithm.Recommended based on the three existing recommendation algorithms are based on the textual content(like product description,rating and user information etc.)mostly for online commodities but many other commodity information cannot be described well with texts,product information is more than just textual contents,images and other visual effects on the effect of commercial users should not be underestimated.To solve the existing problem,this paper will divide commodities' characteristics into category attributes,text content and image feature and import two existing recommendation algorithm: textual content-based recommendation algorithm,category attributes-based recommendation algorithm,and raise up image feature-based algorithm.For the textual content-based recommendation algorithm,certain text information will be extracted,TF-IDF and cosine similarity are used when matching different commodities.For the category attribute-based recommendation algorithm,category attribute matrix is built for similarity calculation.For the image feature-based recommendation SIFT feature of image is used as a matching standard,LSH algorithm is improved based on the p-stable distribution,to match a large number of image of high dimension of matching.The result of experimental verification shows that the improved LSH algorithm has optimized the recall rate and error rate,processing time and the length of hash table are the province of optimization of the memory utilization and search efficiency.Based on these recommendation algorithms this paper has combined them together with certain strategy and put forward the content-related recommendation algorithm for online commodities.Then,this paper constructed the basic recommendation system model(CI-LSH)and verified its efficiency throw experiment.With the given MAE and precision value,the proposed CI-LSH model has improved the accuracy and reliability of the recommended results.Finally,this paper built the entire architecture of online commodity recommendation system,then explained the module function and work flow of this system.This paper has a certain significance in both theoretical innovation and practical application,increasing the recommendation of commodity recommendation,and improving the recommendation efficiency and accuracy of the existing recommendation algorithms.