Collaborative Filtering Algorithm And Its Application In Intelligent Speakers

With the development of digital multimedia technology,digital music has been rapidly promoted.There are tens of thousands of songs in major music websites and platform music libraries,greatly enriching people's spiritual pursuits.However,in the face of huge music data,people find it difficult to find songs that match their minds,and they are lost in the music world.Baidu,Google and other search engines are only suitable for users who have clear requirements.In addition,music is a typical item with a long tail,most users only listen to popular songs,and these songs only account for one of the music library.In a small part,most songs are heard by nobody.Therefore,the personalized music recommendation system promptly determines the user's interest and preference for the user without a clear goal,and recommends a song that suits his/her mind.Collaborative filtering as the most popular recommendation algorithm has been consistently pursued by major Internet companies,but it still faces some challenges such as cold start and poor scalability.In view of this situation,this paper deeply studies the current mainstream recommendation algorithm and fully combines the ideas of data mining algorithms and distributed computing,and improves and optimizes the original algorithm to solve or alleviate the above problems.Firstly,this paper improves and optimizes the recommendation algorithm based on Alternate Least Squares(ALS),uses Canopy-Kmeans algorithm to group users according to user attribute information,and proposes a collaborative filtering recommendation algorithm for grouping,which effectively mitigates the cold start problem of users.The quality of recommendations has also improved.Then,based on the GraphX graph processing framework of Spark platform,the user-item relationship matrix is transformed into a bipartite graph,and the parallel matrix graph calculation is used to rotate the user matrix and the item matrix in the ALS model.Finally,the ALS algorithm is implemented in parallel.The model shortens the running time of the algorithm.After the model is trained,multiple sets of experiments are performed based on the MovieLens data set.Comparing the experimental results,it can be found that the accuracy of the ALS recommendation algorithm model after grouping is improved compared to the traditional ALS recommendation algorithm.In addition,compared with stand-alone environment,even if it is a single node,the execution time of ALS collaborative filtering recommendation algorithm using GraphX is much less than that,and the execution efficiency of the model is improved.Finally,the trained model is used to generate the recommendation result and pushed to the smart speaker's music recommendation module through the embedded database SQLite to achieve personalized recommendation.