"Recommendation System"--Knowledge based recommendation

"Recommender System an Introduction", fourth chapter, based on the recommendation of knowledge.

Overview

The main advantage of collaborative filtering and content-based recommendations is that this knowledge can be acquired and maintained at a relatively small cost.

However, in some scenarios, such as housing, automobiles, computers and other commodities, the synergy system will be poor due to the lack of scoring data, or the time span factor is also important, user preferences with various situations change, etc. these scenarios are not collaborative filtering and content-based recommendations are good at solving.

The knowledge-based Recommender system solves these problems, it does not require scoring data, so there is no startup problem. Based on the knowledge recommendation, the interaction is very strong, so it is a conversational system. Based on knowledge recommendation is not only a filtering system, but more broadly "in a personalized way to guide users to find interesting or useful items in a large number of potential candidates, or produce these items as output results" system.

The two basic types of knowledge-based recommender systems are: constraint-based recommendation and case-based recommendation.

Their referral process is similar: the user must specify the requirements, and then the system tries to give the solution. If a solution is not found, the user must modify the requirements. In addition, the system should also give the explanation of the recommended items.

They differ in how they use the knowledge provided. The case-based recommendation system focuses on retrieving similar items based on different similarity measurements, and the constraint-based recommendation system relies on a well-defined set of recommended rules.

Knowledge representation and Reasoning

Knowledge-based systems rely on detailed knowledge of item characteristics.

Constraint-based recommendation issues can generally be expressed as constraints that are addressed by the constraint Solver, or as a form of a collection of queries that are executed and resolved through the database engine. The case-based recommendation system mainly uses the similarity measure standard to retrieve items from the catalogue.

Constraints

Constraint satisfaction problem (CSP), based on the CSP algorithm and the recommended knowledge base, can be used to construct a constrained recommendation system.

The constraint-based recommender system involves the following variables and constraints:

(1) User attributes (Vc): Describe potential user needs

(2) Product attributes (Vprod): Description of product properties by category

(3) Consistency constraint (CR): Defines the scale of user attributes within the allowable range

(4) filter condition (CF): Defines which product should be selected under which conditions

(5) Product constraints (Cprod): Defines the current valid product classification

(6) Collect query: Unlike the above, a query is a database query that connects a set of selection criteria to a collection method

Example and similarity degree

The case-based recommendation method uses similarity to retrieve items, and the similarity can describe the degree of matching between item attributes and some given user requirements. This similarity calculation, according to different scenarios, has some more general formulas.

Interacting with constraint-based recommender systems

Session Interaction Process

(1) Users to specify their own initial preferences

(2) When sufficient information about the user's needs and preferences is collected, it is provided to the user with a set of matching products, and the user can choose to ask the system to explain why a product is recommended

(3) Users may modify their own needs

Technology to help you interact

These techniques, which help to recommend systems and user interaction, help to improve the usability of the application and achieve higher user acceptance in terms of trust, satisfaction, and results of the recommendations.

(1) Default setting: Recommend default, select Next question

(2) Handling unsatisfactory requirements and empty result sets

(3) Proposed changes to the unmet needs

(4) Sorting based on item/utility recommendation results: It is important to sort the user's utility according to the item, because the first effect, the user will pay more attention to and select items at the beginning of the list, which will significantly improve the trust of the recommended application and the user's willingness to buy. In the knowledge-based conversational recommender system, the ordering of items is based on the multi-attribute utility theory, which is evaluated according to the utility of each item to the user.

Summary of user and constraint-based referral interactions

Users can resort to decision results and modifications when they cannot find a solution.

The default value can give a reasonable candidate scheme to help clarify requirements, and its negative effect is to misuse the default value to manipulate the user.

Based on utility sequencing, it helps to sort the information units on the results page, modify the candidate schemes provided by the decision results and modify the ingredients, and sort the recommended items for interpretation.

These concepts form a toolset that helps to better implement constraint-based recommended applications.

Interacting with instance-based referral systems

Similar to constraint-based recommendations, the early-morning case-based recommendation is a purely query-based approach, where users need to specify (often repeatedly) their needs and know where to find objects. For non-professional knowledge, it is difficult to understand the professional attributes of items, and based on this, people have proposed a browse-based method to retrieve items. Suppose the user does not know what they are looking for, this method navigates for it. Evaluation is a very effective navigation method, and is also a key concept based on the case recommendation system.

(1) Evaluation: The basic idea is that users to the current items to be audited (input items or recommended items) unmet goals to indicate their modification requirements

(2) Mixed evaluation

(3) Dynamic evaluation

(4) Advanced Item recommendation method

(5) Evaluation of diversity

This recommendation process also has a lot of calculation formulas, the calculation process in the inside, when used can be studied in depth.

This recommended process is actually very much like the product retrieval process.

Summary

Based on knowledge recommendation, the quality of the recommended application depends on the quality of the underlying knowledge base, and the process relies on user feedback. Then there are two types of:

(1) Constraint-based recommendations: constraint rules when emphasizing recommendations

(2) Case-based recommendations: Emphasize the results based on the user's initial search results, as well as the gradual evaluation (which can be understood as the choice of more query criteria). Actually very much like conditional retrieval.

"Recommendation System"--Knowledge based recommendation