Construction And Application Of User Portrait Based On Big Data Of Smart Speakers

Smart speakers are based on artificial intelligence and use voice instead of clicking as the main human-computer interaction method.It is a new generation of Internet hardware products.With the rapid increase in the number of users of smart speakers,companies have acquired a large amount of user interaction data with speakers.This article builds smart speaker user portraits based on massive user interaction data with speakers,and introduces the system architecture and applications of user portraits.The construction of the user portrait of the smart speaker firstly analyzes the user-related attributes,and divides the user portrait attributes of the smart speaker into user natural attributes,life cycle attributes,behavior index attributes and user skill attributes according to business requirements.Analyze the attributes,further divide the attributes into labels,and propose three label generation methods based on statistical generation,rule-based generation and algorithm-based generation.User activity and user music value are two typical rule labels.User activity is based on the number of recent interactions with the speaker,artificially set thresholds,and divide users into four categories.The user music value tag represents the commercial value of music skills.This paper proposes to apply the RFM(Customer Relationship Management)model to the construction of users' music value labels,and innovatively redefine R-value,F-value,and M-value.Finally,users are divided into eight categories according to their music value.User family category label and user potential skill label are two typical algorithmic labels.The user's family category tag is to judge whether there are children in the user's family.In terms of feature engineering,this paper innovatively proposes a construction method that combines voiceprint results with resource-extracted features.Compared with the original resource-based feature groups,the model effect has been significantly improved.In order to further improve the predictive ability of the model,the integrated learning stacking method is used to try a combination of different classification models and compare with a single model.Finally,the prediction model using the stacking method has a significant improvement in evaluation indicators.The user potential skill label is a prediction of the user's potential skill.Based on historical data,it is proposed to apply the FP-growth algorithm to the extraction of speaker skills association rules,and obtain user potential skill sets according to the association rules.The system architecture of the user portrait is the overall architecture from the data source to the portrait and then to the application layer,showing the source of the portrait and the application framework.The applications of user portraits include user extraction tools and commercial applications.In order to facilitate the use of user portraits by business personnel,a user extraction tool is constructed based on user portrait data.Commercial applications of user portraits include user insight,report support,precision marketing,refined operations,data research,etc.Specific commercial applications prove the important value of user portraits.