Research on Optimization of Library Management System Based on Hybrid Collaborative Filtering

In order to optimize the personalized recommendation of books, a hybrid-based book resource recommendation method is proposed. Firstly, the method utilizes the global score item between users to improve the similarity calculation method based on Jaccard coefficient, which avoids the situation that the algorithm only considering the user score collection and ignoring the score content cause the misjudgment of the user's preference, and the high similarity between users causes no book to be recommended. Then, the similarity calculation method based on improved Jaccard coefficient is combined with the user-based collaborative filtering recommendation algorithm to generate the user score prediction results, so that the model can still maintain the recommendation performance in the case of the data sparseness. After that, the linear regression model is adopted to mine the user history score data and the item history score data, and then a user score prediction results are generated from the perspective of users and items respectively, so that the model can learn the relationship between users and items more comprehensively. Finally, the score results of the three models are linearly weighted using a hybrid strategy, and users' final score prediction results are obtain. According to the users' predicted scores for different items, personalized recommendation of book resources is achieved. Experiments show that compared with the collaborative filtering recommendation algorithm based on Jaccard coefficient, the collaborative filtering recommendation algorithm based on improved Jaccard coefficient and the improved collaborative filtering recommendation algorithm based on penalty coefficient, the proposed hybrid recommendation algorithm can grasp user preferences more accurately. In the evaluation using the F1-measure method, the F1 value of the proposed algorithm is always the highest among the four algorithms under the condition that the similarity threshold is the same as the number of nearest neighbor users. When the similarity thresholds are 0.4 and 0.5, the F1-Measure of the proposed algorithm is 0.153, 0.330, as well as 0.166, 0.378 and 0.125, 0.102 higher than that of the other three algorithms. In the evaluation using RMSE method, the root-mean-square error of the proposed algorithm remains the smallest among the three errors as the number of the nearest neighbor user increases. When the number of users is 100, the error result is only 0.87, which means the proposes method has more advantages in achieving accurate book recommendation, and proves the effectiveness of the proposed weighted hybrid method in improving the model performance.