Research On Improvement Of Recommendation Algorithm Based On Similarity And Score Prediction

The recommendation algorithm is the most effective solution to the information overload caused by a large amount of information.By recording the user's operational behavior,it is calculated to estimate the information that the user may be interested in.Collaborative filtering recommendation algorithm among many recommendation algorithms is currently the most widely used recommendation algorithm,but there are also problems such as data sparsity,scalability,recommendation accuracy,etc.This paper is based on guided similarity recommendation algorithm and based on neighbor estimation score.In the problem of insufficient recommendation accuracy in the collaborative filtering algorithm,two improved algorithm models are proposed,and the combination of the two algorithms is implemented in the experimental platform.The specific research contents are as follows:1.Through the in-depth study of the principle and its steps of the guided similarity measure model algorithm,a new method is proposed for this algorithm in which the accuracy of the saliency calculation and the dispersion calculation are inaccurate and the recommendation accuracy is reduced.The differential evaluation index algorithm model(DEI)improves the negative impact of the significance and discreteness of the original algorithm model on the accuracy of the recommendation by combining the difference value of the score and the similarity of the interest preference.The algorithm model measures the similarity of the item's scoring coefficient by comparing the scoring value and the scoring value of the same item.Secondly,the user's interest preference difference is analyzed in the overall data set.Finally,the common scoring between users is also considered.The effect of the proportion of the project on the calculation of similarity.The jaccard coefficient is used to reduce the influence of these factors.The MAE values are compared by multiple sets of experimental data and multiple recommended model algorithms.The experimental results show that the DEI algorithm model has higher accuracy of user similarity and lowers to some extent.The impact of data sparseness on the quality of recommended information.2.Through the in-depth study of the principle and its steps of the Nearest Neighbor Estimation Scoring Model algorithm,and this algorithm does not take into account the similarity between the unscoring items and the users already scored items when calculating the estimated score.Aiming at the problem of accurate accuracy reduction of estimated scores,a new algorithm model based on item similarity formation and estimation score is proposed(IFSP).The algorithm model builds a scoring matrix by similar clustering of items,waiting for the estimated items and users.The number of similar items has been scored as the weight of the item,combined with the similarity of the specific user the estimated score is reasonably debugged,the estimated scoring mode is optimized,and the MAE values are compared through multiple sets of experimental data and multiple scoring models.The results demonstrate that the IFSP predictive scoring method significantly improves the accuracy of the estimated score.3.Based on the improved DEI algorithm model and IFSP algorithm model,the integration of two improved algorithm models is realized by constructing the experimental platform using the spark2.0 calculation framework.By testing under multiple sets of sample data sets,it is proved that the integrated algorithm model is In practical applications,the efficiency of the algorithm can be effectively improved,thereby improving the scalability of the algorithm.