Methods For Similarity Calculation In Personalized Recommendation

In recent years,various types of data in platforms such as e-commerce websites have become more and more concentrated.Using these auxiliary information to im-prove the accuracy of recommendation systems has become one of the trends in the development of recommendation systems.Similarity calculation is the most common method of recommendation with auxiliary information,but existing systems often only consider the similarity measure between objects of the same type,so that the implicit semantic information in the auxiliary information is not fully utilized,resulting in a decline in the service quality of the recommended system.Therefore,the similarity calculation algorithm that fully exploits implicit semantic information has become a research hotspot.In this thesis,we focus on the recommendation system under the heterogeneous information network scenario.For the problem of low utilization of implicit semantic information in auxiliary information,the goal of improving the recommendation accu-racy of the personalized recommendation system is to take the random walk algorithm as the core method.The main work of the similarity calculation algorithm for person-alized recommendation is as follows:(1)In order to mine the implicit semantics of different nodes on the asymmetric path in the heterogeneous information network,this thesis proposes a new walk algo-rithm under the path constraint-"Asynchronous bidirectional random walk algorithm"is calculated according to this walk algorithm.The similarity between different object types,and the implicit semantics of all nodes on the path are obtained.At the same time,the recommendation algorithm based on this walk mode is used to verify the ex-periment on the actual data set.The result shows that the recommended results using this kind of walk algorithm are better than other similar methods.(2)In order to use the explicit feedback data such as scoring to accurately describe the user's preferences,this thesis proposes an algorithm that can calculate the similar-ity in the system of user explicit feedback and implicit feedback.The algorithm first sets the walk probability according to the score size by taking a hop with a score or the like on the walk path,and then uses the asynchronous bidirectional random walk algorithm to perform the similarity calculation.At the same time,considering that the data between one hop includes not only explicit feedback but also implicit feedback,the algorithm first uses implicit feedback data to predict the score,and then combines the predicted score with the original display feedback data to calculate the similarity.Finally,the thesis evaluates the recommended quality and error rate of the algorithm on the IMDb dataset.The results show that the target of higher recommendation quality and lower error is achieved.