Research On The Core Algorithm Of Multi-scenario Recommendation System

Recommender system can be active service,personalized customization,and so on,therefore it is widely used in e-commerce,online video,online music,sequence recommender and other fields.However,when recommender systems are applied to various fields,they are faced with different recommendation scenarios.Their core algorithms are also quite different.In the many of scenarios of recommender system,this paper studies two representative recommendation scenarios,preference recommendation and temporal recommendation.Preference recommendation scenarios are common in information services such as e-commerce and online entertainment services.It establishes a user's interest model based on the user's historical data,predicts the user's preference,and makes the recommendation by the corresponding strategy.Compared with the preference recommendation scenario,the temporal recommendation scenario pays more attention to the influence of time factors,establishes the time sequence user interest model,and gives the timely recommendation.It is common in areas where information will change over time,such as sales forecast,security check,fault warning,etc.First of all,this paper proposes two new similarities,namely,Triangle similarity and TMJ similarity,for the kNN-based collaborative filtering recommendation algorithm under the preference recommendation scenarios.The two rating vectors can construct a triangle in the three-dimensional space.According to the triangle relation,the Triangle similarity is proposed in this paper.It takes into account the vectors' length and angle.However,it only takes into account the information of the common rated project.In this paper,we proposed TMJ similarity,which integrating Triangle similarity and Jaccard similarity.We compare the new similarity measure with eight state-of-the-art ones on four popular datasets.Results show that the MAE and RSME obtained by using TMJ similarity are better than other similarity,and the performance of using Triangle similarity is also better than most similarity.Secondly,under the time series recommendation scenario,we proposed an interactive sequential recommendation algorithm based on mobile average prediction,which is in the background of safe check of drilling workers.Two techniques are used to the recommendation algorithm.The first technique is the time series prediction.In the initial stage,the moving average method used to predict the number of occurrences of all the items to check.The second technique is user interaction technology,which used to adjust subsequent recommended subsequences after the interaction between the system and the user.Experiments on a real-world dataset show compared with the existing detection schemes,the algorithm provides a better detection sequence and reduces the security risk.