The recommendation method is the most core and critical part of the entire recommendation system, which largely determines the performance of the recommendation system. At present, the main recommendation methods include: recommendation based on content, recommendation based on collaborative filtering, recommendation based on association rules, recommendation based on utility, recommendation based on knowledge and recommendation based on combination.
1. Content-based recommendation
Content-based recommendation (Content-based Recommendation) is the continuation and development of information filtering technology. It is based on the content information of the project to make recommendations, and does not need to be based on the user's evaluation of the project, and more needs to use machine learning The method obtains the user's interest profile from the instances of the feature description about the content. In a content-based recommendation system, an item or object is defined by the attributes of related features. The system learns the user's interest based on the characteristics of the user's evaluation of the object, and examines the degree of matching between the user's profile and the item to be predicted. The user's data model depends on the learning method used. Commonly used are decision trees, neural networks and vector-based representation methods. Content-based user profiles require historical data of users, and the user profile model may change as the user's preferences change.
The advantages of content-based recommendation methods are:
1) No need for other users' data, no cold start problem and sparse problem.
2) It can be recommended for users with special interests.
3) Can recommend new or not very popular items, no new item issues.
4) By listing the content characteristics of recommended items, it is possible to explain why those items are recommended.
5) There are relatively good technologies, such as the technology on classification learning is quite mature.
The disadvantage is that the content can be easily extracted into meaningful features, the feature content is required to have good structure, and the user's taste must be expressed in the form of content features, and the judgment of other users cannot be explicitly obtained.
2. Collaborative filtering recommendation
Collaborative Filtering Recommendation (Collaborative Filtering Recommendation) technology is one of the earliest and most successful technologies used in recommender systems. It generally uses the nearest neighbor technology, uses the user's historical preference information to calculate the distance between users, and then uses the weighted evaluation value of the target user's nearest neighbor users to evaluate the product to predict the target user's preference for a specific product. A preference level to recommend to target users. The biggest advantage of collaborative filtering is that there are no special requirements for recommended objects, and it can deal with unstructured and complex objects, such as music and movies.
Collaborative filtering is based on the assumption that a good way for a user to find what he is really interested in is to first find other users who have similar interests to this user, and then recommend the content they are interested in to this user. Its basic idea is very easy to understand. In daily life, we often use the recommendations of good friends to make some choices. Collaborative filtering applies this idea to the e-commerce recommendation system, and recommends to target users based on the evaluation of other users on a certain content.
The recommendation system based on collaborative filtering can be said to make corresponding recommendations from the user's point of view, and it is automatic, that is, the recommendation obtained by the user is implicitly obtained by the system from the purchase mode or browsing behavior, etc. Recommended information about your own interests, such as filling out some survey forms, etc.
Compared with content-based filtering methods, collaborative filtering has the following advantages:
1) Ability to filter information that is difficult to perform automatic content analysis by machines, such as artwork, music, etc.
2) Share the experience of others, avoid incomplete and imprecise content analysis, and be able to filter based on some complex, difficult to articulate concepts (such as information quality, personal taste).
3) Ability to recommend new information. Information that is completely dissimilar in content can be found, and the content of the recommended information cannot be expected by the user in advance. This is also a big difference between collaborative filtering and content-based filtering. Many content-based filtering recommendations are content that users are already familiar with, while collaborative filtering can discover users’ potential but undiscovered interests and preferences.
4) Can effectively use the feedback information of other similar users, reduce the amount of user feedback, and speed up the speed of personalized learning.
Although collaborative filtering has its considerable application as a typical recommendation technology, there are still many problems to be solved in collaborative filtering. The most typical problems are Sparsity and Scalability.
3. Recommendation based on association rules
Association rule-based recommendation is based on association rules, with the purchased product as the rule header and the rule body as the recommendation object. Association rule mining can find the correlation of different commodities in the sales process, and has been successfully applied in the retail industry. The management rule is to count in a transaction database what percentage of the transactions that purchased commodity set X purchase commodity set Y at the same time. The intuitive meaning is how much users tend to buy other commodities when they purchase certain commodities. . For example, when buying milk, many people buy bread at the same time.
The discovery of association rules in the first step of the algorithm is the most critical and time-consuming, and is the bottleneck of the algorithm, but it can be done offline. Secondly, the problem of synonymy of commodity names is also a difficulty of association rules.
4. Utility-based recommendation
Utility-based recommendation is calculated based on the utility of users using items. The core problem is how to create a utility function for each user. Therefore, the user profile model is largely based on It is determined by the utility function adopted by the system. The benefit of utility-based recommendation is that it can take non-product attributes such as Vendor Reliability and Product Availability into account in utility calculations.
5. Recommendation based on knowledge
Knowledge-based recommendation can be regarded as an inference technology to a certain extent, and it is not recommended based on user needs and preferences. Knowledge-based methods are clearly distinguished by the functional knowledge they use. Functional knowledge is knowledge about how an item satisfies a particular user, and thus explains the relationship between needs and recommendations, so user profiles can be any knowledge structure that supports reasoning, and it can be normalized by users. A query can also be a more detailed representation of the user's needs.
6. Combination recommendation
Since various recommendation methods have advantages and disadvantages, in practice, Hybrid Recommendation is often used. The most researched and applied is the combination of content recommendation and collaborative filtering recommendation. The simplest way is to use the content-based method and the collaborative filtering recommendation method to generate a recommendation prediction result respectively, and then combine the results by a certain method. Although there are many recommended combination methods in theory, they are not all effective in a specific problem. One of the most important principles of combined recommendation is to avoid or make up for the weaknesses of their respective recommendation technologies after combination.
In terms of combination methods, some researchers have proposed seven combination ideas:
1) Weight : Weight the results of multiple recommendation techniques.
2) Switch : According to the background of the problem and the actual situation or requirements, it is decided to adopt different recommendation techniques for switching.
3) Mixed : At the same time, multiple recommendation technologies are used to give multiple recommendation results to provide reference for users.
4) Feature combination : Combining features from different recommendation data sources is used by another recommendation algorithm.
5) Cascade : First, a recommendation technique is used to generate a rough recommendation result, and the second recommendation technique further makes a more accurate recommendation based on the recommendation result.
6) Feature augmentation : One technique produces additional feature information embedded into the feature input of another recommendation technique.
7) Meta-level : The model produced by one recommendation method is used as the input of another recommendation method.