| With the rapid development of the Internet,MOBA games have become one of the most popular games in the world by virtue of their fun,competitive and social properties,and have formed a unique user community and e-sports industry.Items are a key factor in MOBA games,and different item choices can provide users with different game experiences and rich tactical options.Items in MOBA games mainly include heroes and equipment.In a match,the user needs to select the hero in the hero selection stage,and select the equipment for the hero in the match stage.As the number of games increases,users will develop their own personalized hero preferences.The equipment has different attributes,and has matching or controlling relationships with different heroes.Each hero has a set of equipment that goes with it,and this set of equipment has to be adjusted as the game changes.Therefore,the item recommendation needs to fully explore the user’s personalized preferences and the context of the game.MOBA games have a large number of items and a huge number of items combinations,making it difficult for ordinary users to make appropriate choices.Developing an item recommendation system for the MOBA game scene can improve the user’s game experience.This paper mainly studies the problem of item selection in different stages of MOBA games,and designs and implements an item recommendation system for MOBA game scenes.The main contents are as follows:(1)A personalized MOBA hero recommendation model PHPRec based on attention mechanism and noise filtering is proposed.This model is used to predict the user’s hero preference before the hero selection stage.It uses the attention mechanism to mine the rich performance information in the user’s historical match data,obtains the correlation between the user’s performance and the user’s interest,and introduces the filter layer to filter the noise data that may exist in the user’s historical match sequence.Experimental results on real datasets show that PHPRec is superior to some common recommendation algorithms in personalized hero recommendation scenarios.(2)A context-aware MOBA hero recommendation model GHPRec is proposed,which integrates users’ personalized preferences.This model is used to recommend appropriate heroes to the user during the hero selection phase.Considering the limitations of the hero selection process and game rules in the hero selection stage of MOBA games,it is unreasonable to select heroes only based on users’ personalized hero preferences.In this paper,the user behavior network is designed to model the user’s personalized hero preference,and the match network is designed to integrate the user’s personalized hero preference in the match stage and perceive the context information of the match stage.Based on this,the hero recommendation is made.Experiments on real MOBA game datasets show that GHPRec effectively improves HR and NDCG.(3)A MOBA equipment recommendation model GEPRec based on hero equipment pattern analysis and context awareness is proposed.This model is used to recommend equipment for heroes after the hero selection stage and before the match stage.It includes two modules based on Transformer encoder:the hero equipment mode analysis module analyzes the relationship between heroes and equipment according to the hero’s historical match-up;The match module merges the hero’s equipment mode with the context information of the current match,and then recommends equipment.Experimental results on LOL and Dota 2 datasets show that GEPRec achieves excellent results.(4)A sequential MOBA equipment recommendation model GSEPRec based on context information awareness is proposed.This model is used to recommend the next equipment to the hero in order during the match phase.Considering the order of the equipment selection in a game,the order encoder is designed to capture the order information in the equipment purchase behavior.The context encoder is used for modeling the interaction relationship between heroes and their equipment in a match.Based on this,the next equipment is recommended.Experiments on real datasets show that the GSEPRec model improves the recommendation effect.(5)MOBA games have a large market size and a large number of users,but there is a lack of personalized MOB A item recommendation system in the domestic market.Therefore,this paper designs and implements an item recommendation system for MOBA game scene.The models proposed in this paper are applied to the system,which can provide users with hero recommendation,lineup recommendation,equipment recommendation and item details,so as to improve user experience. |
