A hybrid collaborative filtering model based on depth learning in Recommendation System

Model collaborative

CSDNyunjisuan· 2017-02-04 23:51:09

in recent years, deep learning breakthrough and great achievements have been achieved in speech recognition, image processing, Natural Language Processing etc.. Relatively speaking, the research and application of deep learning in the recommendation system is still in the early stage.

Ctrip in the combination of depth and learning recommendation system also involves the relevant research and application, and published research results of the corresponding A Hybrid Collaborative "Filtering Model with Deep Structure for Recommender Systems" in the International Conference on artificial intelligence AAAI 2017, this paper will share a deep learning application in recommender system, and introduces the practice of Ctrip based BI team in this field.

," box-sizing: style= recommendation system introduced

recommended the function of the system is to help users find the initiative to satisfy their preferences personalized items and recommended to the user. The input data of the recommendation system can be varied, which can be divided into three levels: User, Item and Ratings, respectively. For a particular user, the output of the recommender system is a list of recommendations that give the user a list of items that may be of interest.




as shown on the right in Figure 1, describe the recommendation problem of a typical form is as follows: we have a large sparse matrix, each row of the matrix represents a User, each column represents a Item, each "+ matrix" The number of User on the Item of the Rating, which can be a value of two values, like and do not like; can also be 0~5 scores, etc..

is inferred every User for every Item, "recommended" to solve the problem according to the results of the calculation and prediction step recommended he did not play too much Item to the user. But at present most recommendation algorithms concentrate on the "prediction" link, "recommended" link according to the prediction process of calculating the scores in descending order according to the recommended to the user, this scheme is mainly introduced to share "missing" prediction score matrix in R score.

two, based on collaborative filtering recommendation

< P style= box-sizing: border-box margin-bottom: 20px; color:; RGB (51, 51, 51);;;;;;;" > collaborative filtering recommendation based on user behavior by collecting the past to get the goods on display or implicit information, according to user preferences on items, or items that correlation between users, and then recommend these relations based on.  




Memory-based method recommended by performing a nearest neighbor search, each Item or User as a vector similarity calculation for all other Item or User with it. With Item or User between two
The lastest articles of CSDNyunjisuan

A hybrid collaborative filtering model based on depth learning in...

AndroidImageView correct use of posture

Using OpenStack to build a high speed cloud platform

With this story to explain the CAP, is so simple!

Interview with PaulConroy Asia Pacific Business Manager Atlassian: more than...

Jingdong: Jingdong cloud cloud chief architect Yang Haiming quest, to service...