|
| Recommended algorithm classification | Representative algorithm | Main advantage | Main disadvantage |
|
| Content-based recommendations | Text recommendation method | (1) It can handle the cold start problem of new projects
(2) Solve the problem of sparse scoring
(3) Be able to make recommendations for users with special interest
(4) Can recommend new products and nonpopular products
(5) The reason for the recommendation can be explained intuitively | (1) Unstructured data cannot be effectively processed
(2) Lack of potential mining capabilities for users
(3) There is a cold start problem when a new user appears
(4)Poor self-learning ability |
| Memory-based collaborative recommendation | (1) Recommendations based on neighbors
(2) Based on user/project Top-N recommendation | (1) Suitable for processing complex unstructured data
(2) Effectively discover projects with high novelty
(3) Recommended explanations can be provided | (1) There is a cold start problem
(2) When the data scale expands and the data sparseness is severe, the quality and efficiency of the recommendation will decrease
(3) Rely on historical behavior data |
| Model-based collaborative recommendation | (1) Bayesian network
(2) Clustering (k-means, neural network)
(3)Dimensionality reduction technology (SVD, LDA, pISA, PCA, etc.)
(4) Graphical model | (1) Effectively alleviate the problem of data sparsity
(2) Enhance the scalability of the system
(3) Improved prediction accuracy | (1) The modeling is complex and needs to be updated periodically
(2) Dimensionality reduction will cause some information loss, and it is difficult to provide recommended explanations |
| Knowledge-based recommendation | (1) Recommendation based on constraints
(2) Recommendations based on examples
(3) Recommendation based on knowledge reasoning | (1) Do not rely on the user’s historical behavior data
(2) Solve the cold start problem | (1) The quality of recommendation depends on knowledge acquisition and quality
(2) Recommendations are static |
| Hybrid collaborative filtering | (1) Content-based hybrid collaborative filtering
(2) Memory-based and model-based hybrid collaborative filtering | (1) Mixed recommendations can learn from each other’s strengths
(2) Use content analysis to deal with new projects or new user issues | (1) Method selection and combination sequence
(2) The weight distribution of the results obtained by different methods |
|
