CN-122019598-A - Intelligent correction selecting method

Abstract

An intelligent correction selecting method comprises the following steps of 1, collecting correction selecting parameters given in a mode of structuring a form or natural language or a mixture of the two, 2, carrying out standardization and fault tolerance verification processing on the correction selecting parameters, then automatically judging a current recommended scene according to a pre-built unified condition parameter object model or adding the current recommended scene into the pre-built unified condition parameter object model, 3, activating a rule engine, dynamically analyzing the correction selecting parameters through a configured expression, 4, generating unique query keywords based on the pre-built unified condition parameter object model, searching corresponding recommended results in a Redis cache through the keywords, and 5, if no corresponding recommended results exist in the Redis cache, constructing a dynamic query statement according to the rule engine to query the recommended results.

Inventors

• LI YAN

Assignees

• 北京小希教育科技有限公司

Dates

Publication Date

20260512

Application Date

20251105

Claims (4)

1. 1. An intelligent correction selecting method is characterized by comprising the following steps: Step 1, collecting selected parameters given in a mode of a structural form or natural language or a mixture of the structural form and the natural language; Step 2, carrying out standardization and fault tolerance verification processing on the selected calibration parameters, and then automatically judging the current recommended scene according to a pre-constructed unified condition parameter object model or adding the recommended scene into the pre-constructed unified condition parameter object model; step 3, activating a rule engine, and dynamically analyzing the correction parameter through a configured expression; step 4, generating a unique query keyword based on the pre-constructed uniform condition parameter object model, and searching a corresponding recommendation result in a Redis cache through the keyword; And step 5, if no corresponding recommendation result exists in the Redis cache, constructing a dynamic query statement according to the rule engine to query the recommendation result.
2. 2. The intelligent calibration method according to claim 1, wherein pre-calculation analysis is performed daily in the Redis cache according to the query keyword so as to store the calculated recommendation result in advance.
3. 3. The intelligent calibration method according to claim 2, further comprising step 6 of processing the recommended results and forming desired output results.
4. 4. The intelligent school selecting method according to claim 3, wherein the step 6 includes a step 6-1 of generating a label hint for each recommended result, a step 6-2 of eliminating the institutions which do not accord with the school selecting parameters or lack of the support of the recorded data, a step 6-3 of carrying out weighted scoring according to the characteristics of the users and sorting according to the weighted scoring, and a step 6-4 of selecting the most scored preset number of institutions.

Description

Intelligent correction selecting method Technical Field The invention relates to the field of intelligent reasoning of a knowledge base, in particular to an intelligent correction selecting method. Background In the current reservation and delivery service business, the agent selection and correction process often faces the technical problems of high parameter dimension and complex combination, wherein the personalized background of the user comprises achievements, language capability, professional intention, application learning period, target country and the like, thousands of possible combinations are formed, and the traditional manual processing cannot be matched accurately. The rule logic is frequently changed and difficult to maintain, the recording standards of different countries and different projects are frequently updated, and the traditional code type logic is difficult to quickly respond to the rule change. The searching efficiency is bottleneck, namely multi-condition combined searching is carried out in the structured database, complex SQL inquiry is easy to form, and the response performance of the system is seriously affected if no optimization exists. The result interpretation is insufficient, that is, the returned result of most recommendation systems is opaque, and the explanation of why the institution is recommended cannot be clearly performed, so that the user trust degree is reduced. The result lacks diversity and dynamics, and under the condition of lacking a post-processing mechanism, the search result can be seriously homogenized, so that the breadth and the precision are difficult to be combined. Disclosure of Invention Therefore, the invention provides an intelligent correction selecting method, which comprises the following steps: Step 1, collecting selected parameters given in a mode of a structural form or natural language or a mixture of the structural form and the natural language; Step 2, carrying out standardization and fault tolerance verification processing on the selected calibration parameters, and then automatically judging the current recommended scene according to a pre-constructed unified condition parameter object model or adding the recommended scene into the pre-constructed unified condition parameter object model; step 3, activating a rule engine, and dynamically analyzing the correction parameter through a configured expression; step 4, generating a unique query keyword based on the pre-constructed uniform condition parameter object model, and searching a corresponding recommendation result in a Redis cache through the keyword; And step 5, if no corresponding recommendation result exists in the Redis cache, constructing a dynamic query statement according to the rule engine to query the recommendation result. Preferably, pre-calculation analysis is performed daily in the Redis cache according to the query keyword so as to store the calculated recommendation result in advance. Preferably, the method further comprises a step 6 of processing the recommended result and forming a desired output result. Preferably, the step 6 includes a step 6-1 of generating a label hint for each recommended result, a step 6-2 of eliminating institutions that do not match the selected parameters or lack the support for the recorded data, a step 6-3 of weighting and ranking the selected institutions according to the characteristics of the user, and a step 6-4 of selecting the most highly scored predetermined number of institutions. In particular, for the break-up algorithm, the following is detailed: In actual "smart calibration" recommendations, the original results queried by the database are most likely to be filled by a particular specialty (e.g., master of Business Analytics) or country (e.g., the united states). This can lead to the following problems: (1) The recommendation list "looks the same" and the first page will show the first few solutions, which if not controlled, would be likely to see the profession of the first few solutions the same. (2) If the profession of the United states is more in line with the user's condition, the display is probably mainly based on the United states, and the profession of other countries is delayed (3) Single recommendation, lack of explorability (which is very bad in leave-on business). What are the more diverse leave-on scenarios the user wants, such as what country i can go to leave? In practical application, firstly dimension selection is performed, and a certain breaking dimension (country and specialty) is selected preferentially according to service requirements. Then, a barrel dividing operation is performed to divide the original recommendation result into a plurality of barrels, and schools in the same country (or the same professional) are in each barrel. For example, the original result is 100, and is divided into 5 national barrels, namely the United states, the United kingdom, the Australia, the Canada and the Singapore. T