Collaborative filtering method based on local optimization dimension reduction and clustering
A local optimization and collaborative filtering technology, which is applied in computer parts, character and pattern recognition, special data processing applications, etc., can solve the problem of inability to balance recommendation time and recommendation accuracy
- Summary
- Abstract
- Description
- Claims
- Application Information
AI Technical Summary
Problems solved by technology
Method used
Image
Examples
Embodiment 1
[0059] A collaborative filtering method based on local optimization dimensionality reduction and clustering is characterized in that: first, the sparse user-item rating matrix is subjected to dimensionality reduction processing to obtain a user feature matrix; secondly, clustering techniques are applied to the user feature matrix to obtain clusters of similar users ; Then predict the score of the target user on the user test set; finally select the N items with the highest score according to the prediction results to generate recommendations;
[0060] The collaborative filtering method based on local optimization dimensionality reduction and clustering is characterized in that: constructing the approximate difference matrix of the user-item rating matrix includes the following steps:
[0061] Step 1: The singular value decomposition theorem for local optimization shows that for all matrices C[k,n] where k rows represent users and n columns represent items C can be decomposed ...
Embodiment 2
[0069] The feature of the collaborative filtering method based on local optimization dimensionality reduction and clustering is: based on the local optimization dimensionality reduction method, the learning rate sr 1 When the difference between the mean square error (Mean Square Error, MSE) of the two iterations before and after the iteration is less than the threshold β, use a smaller learning rate sr 2 The iterative local optimization singular value decomposition method includes the following steps:
[0070] Step 1: Initialize
[0071] PMSE=0; Sum=0; sr 1 = 0.003; sr 2 = 0.00005; λ = 0.12; β = 0.0003
[0072] Step 2: For the user item set (i, j) in the training set D:
[0073] (1) Calculate user i's rating on item j:
[0074] (2) Calculate the error between the predicted score and the real score: Sum=r ij r ij ;
[0075] (3) For all features f (1≤f≤s) use the gradient descent method to solve:
[0076] x if =X if -sr 1 (r ij ·X if +λY jf ); Y jf =Y jf -sr...
Embodiment 3
[0089] The described collaborative filtering method based on local optimization dimensionality reduction and clustering is characterized in that: the K-means clustering method comprises the following steps:
[0090] Step 1: randomly select K users as K centroids;
[0091] Step 2: The remaining users are assigned to the nearest cluster according to their distance to each centroid; Pearson similarity is used to calculate the distance value; the similarity sim(i,j) between user i and user j is:
[0092]
[0093] where I ij The set of items rated jointly by user i and user j is C ip Indicates the rating of user i on item p with Respectively represent the average ratings of users i and j on the common rating items;
[0094] Step 3: Calculate the mean of the user class to define a new centroid;
[0095] Step 4: Recalculate the distance for each user to update the cluster to which the user belongs;
[0096] Step 5: Reassign according to the distance between the user and th...
PUM
Abstract
Description
Claims
Application Information
- R&D Engineer
- R&D Manager
- IP Professional
- Industry Leading Data Capabilities
- Powerful AI technology
- Patent DNA Extraction
Browse by: Latest US Patents, China's latest patents, Technical Efficacy Thesaurus, Application Domain, Technology Topic.
© 2024 PatSnap. All rights reserved.Legal|Privacy policy|Modern Slavery Act Transparency Statement|Sitemap