A method for operating a part-time job recommendation system with a fallback mechanism

Abstract

This invention discloses a method for operating a part-time job recommendation system with a fallback mechanism. By setting up a fallback tag library, this invention provides fallback keywords to blank tag pages when the matching and filtering of job postings in the tag pool with user requests fails. These blank tag pages can then display results based on the fallback keywords, thereby improving user experience and ensuring exposure for certain job postings. This effectively utilizes traffic and guarantees the effectiveness of the part-time job recommendation system.

Description

Technical Field

[0001] This invention relates to the field of part-time job recommendation applications, and specifically to a method for operating a part-time job recommendation system with a fallback mechanism. Background Technology

[0002] The recruitment industry is currently undergoing rapid development and transformation. On the one hand, with the emergence of the internet and the wave of mass entrepreneurship, various industries are facing industrial upgrading, and the focus of competition among enterprises is on talent. Therefore, talent recruitment has become a top priority for all companies. Part-time positions fall under the category of flexible employment and are relatively unique. Not only are the types of positions diverse and the work cycles short, but the skill and experience requirements also differ significantly from those of ordinary full-time positions. Therefore, to achieve the best part-time job recommendation results, different part-time job recommendation systems need to be built based on different user characteristics and job characteristics. Currently, when using part-time job recommendation systems, because there are both online and offline types of part-time jobs, and part-time jobs themselves have certain requirements regarding users' age, gender, and education, online jobs are closed by businesses at irregular intervals, and offline jobs have some LBS (Location Based Service) characteristics. This leads to a situation where, after filtering and processing by factors such as user basic information and distance within the tag pool provided by the part-time job recommendation system, no suitable jobs are displayed, resulting in blank pages. Consequently, the relevant recommendation results received by users are blank, which undoubtedly reduces the user experience and leads to low exposure for certain jobs. Summary of the Invention

[0003] The purpose of this invention is to provide a method for operating a part-time job recommendation system with a fallback mechanism. This invention ensures the effectiveness of the part-time job recommendation system, improves user experience, and increases the exposure of certain job postings.

[0004] To solve the above-mentioned technical problems, the present invention provides the following technical solution: A method for operating a part-time job recommendation system with a fallback mechanism, comprising the following steps:

[0005] S1. In the part-time job recommendation system, based on actual business needs, the operations team will tag jobs that meet certain characteristics with the same label and place them in a tag job pool. These jobs will be displayed on the platform or advertised externally via a tag landing page.

[0006] S2. When a user makes a request, the jobs circled in the job pool are matched and filtered against the user's request. After matching and filtering, the remaining jobs that meet the requirements are finely ranked and diversified, and then the results are returned to the user. If the list of remaining jobs after matching and filtering is empty, it is considered that the landing page of the tag is blank. In the case of blank, the keywords of the part-time job description text information with this tag are extracted to form the tag name, and then proceed to step S3.

[0007] S3. Input the tag name requested by the user into the tag catch-all terminology library. Based on the tag name, retrieve the tag-related catch-all keywords and weights from the tag catch-all terminology library. Use the catch-all keywords and weights to search the entire job pool as a catch-all, and then return the results to the user.

[0008] In the above-mentioned method for operating a part-time job recommendation system with a fallback mechanism, step S2 includes matching filtering, which includes filtering based on user basic information, filtering for jobs that have temporarily suspended recruitment, filtering based on distance factors, and filtering based on risk control.

[0009] The aforementioned method for operating a part-time job recommendation system with a fallback mechanism involves refining and diversifying the job list after matching and filtering, sorting it according to different business objectives using a click-through rate model or a conversion rate model to form a refined job list. When selecting the top N jobs after refining, diversification is then performed by interspersing and dispersing jobs of the same brand and purpose.

[0010] The aforementioned method for operating a part-time job recommendation system with fallback processing involves training a word2vec model using text information from a part-time job database to obtain a part-time job similarity word model. Then, using the tag names, multiple similar words are retrieved from the similarity word model, and the similarity between each similar word and the tag name is obtained. This similarity is used as the weight for keyword search matching. This information is then synchronized to a fallback keyword weight configuration table, which is subsequently imported into an Elasticsearch search DSL to form the fallback tag database.

[0011] The aforementioned method for operating a part-time job recommendation system with a fallback mechanism involves manually adding a batch of fallback keywords and weight items to a specific tag's fallback keyword library, which are then synchronized to the fallback keyword weight configuration table as a second-level fallback.

[0012] The aforementioned part-time job recommendation system operation method with fallback processing also includes a fallback tag keyword library with system keywords and weight items for key levels at the system level, which are synchronized to the fallback keyword weight configuration table as the lowest level fallback.

[0013] In the aforementioned method for operating a part-time job recommendation system with fallback processing, step 2 involves extracting keywords using the TextRank algorithm in NLP.

[0014] Compared to existing technologies, this invention establishes a fallback keyword library. This library provides fallback keywords for blank tag pages when the matching of job postings in the tag pool with user requests fails. These blank tag pages can then display results based on these fallback keywords, improving user experience and ensuring exposure for certain job postings. This effectively utilizes traffic and guarantees the effectiveness of the part-time job recommendation system. Furthermore, the use of fallback keywords ensures a certain relevance between user requests and returned job results, avoiding serious mismatches. The fallback keyword library employs a multi-level fallback approach, prioritizing the display of the most relevant keywords. The invention utilizes manual addition, allowing for effective handling of extreme situations where synonyms associated with specific tag names cannot be matched in job postings. Manually increasing the weight of certain keywords further enhances the accuracy of the fallback. For unacceptable blank tag pages during deployment, this invention uses broad system keywords to minimize the blank rate, ensuring a good user experience while increasing traffic to relevant job postings and providing good exposure for businesses. Attached Figure Description

[0015] Figure 1 This is a schematic diagram illustrating the principle and process of the present invention;

[0016] Figure 2 This is a diagram illustrating the multi-level weighted keyword fallback configuration. Detailed Implementation

[0017] The present invention will be further described below with reference to the embodiments and accompanying drawings, but this should not be construed as limiting the present invention.

[0018] Example: A method for operating a part-time job recommendation system with a fallback mechanism, such as... Figure 1 As shown, it includes the following steps:

[0019] S1. In the part-time job recommendation system, based on actual business needs, operations teams will tag jobs that meet certain characteristics with the same label and place them in a tagged job pool. These jobs will then be displayed on the platform or advertised externally via a tagged landing page. (There are many such tagged job pools, each a subset of the overall job pool.) Figure 1 The operations staff sets job tags according to the configuration rules, thereby generating a tag page with an ID of 1.

[0020] S2. When a user makes a request, the job postings circled in the job posting pool are matched and filtered against the user's request, removing those that do not meet the requirements. That is... Figure 1The system matches user requests with the tag with ID 1 for filtering. This filtering includes filtering based on basic user information, filtering for temporarily suspended job postings (referred to as pause filtering), distance factor filtering, and risk control filtering. If the matching filter is successful, the tag with ID 1 is recalled, and the corresponding job postings are refined and diversified (the refined and diversified processing refers to sorting the filtered job postings according to different business objectives using a click-through rate (CTR) or conversion rate (CVR) model to form a refined job posting result. When selecting the top N refined job postings, diversification processing is performed, primarily using the logic of interspersing and dispersing postings of the same brand and category. The diversified job posting list has a richer presentation of brands and categories). The result is then returned to the user. If the remaining job posting list after matching and filtering is empty (the probability of emptiness depends on the size of the job posting pool for this tag), the landing page for that tag is considered blank. In the case of a blank page, the device extracts keywords from the description text of the part-time job postings tagged with that type using the TextRank algorithm of NLP to form tag names, and proceeds to step S3.

[0021] S3. Input the user-requested tag name into the tag catch-all terminology. In this embodiment, the tag catch-all terminology has three levels of catch-all. The first level of catch-all is that the tag catch-all terminology uses the text information in the part-time job database to train a word2vec model (which can be implemented using Spark MLlib, a machine learning algorithm library) to obtain a part-time job similarity word model. Then, the tag name is used to obtain the top N similar words in the similarity word model, and the similarity between each similar word and the tag name is obtained (for example, if the tag name is waiter, the top 2 similar words are waiter and food preparer, with similarities of 0.8 and 0.65 respectively). The similarity is used as the weight for keyword search matching. The above information is synchronized to the catch-all keyword weight configuration table, and then the catch-all keyword weight configuration table is imported into the Elasticsearch search DSL to form the tag catch-all terminology.

[0022] The second level of fallback involves manually adding a batch of fallback keywords and weight items to the fallback keyword library for specific tags, and synchronizing them to the fallback keyword weight configuration table. These fallback keywords and weight items can be set manually.

[0023] The lowest-level catch-all keyword library also includes system-level key terms (e.g., "part-time job," "job") and weightings (e.g., 0.01 and 0.02), which are synchronized to the catch-all keyword weighting configuration table. The lowest-level catch-all can be enabled in certain campaign scenarios to reduce the blank rate of tag pages using system-level key terms. However, the weight of these system-level key terms cannot exceed that of the selected keywords from the second and first-level catch-all categories to avoid recommendation confusion.

[0024] Therefore, as Figure 2 As shown, a multi-level weighted keyword catch-all tag library is obtained. Finally, based on the tag name, the catch-all keywords and weights associated with the tag are obtained from the catch-all tag library. The catch-all keywords and weights are then used to search the entire job pool. It should be noted that the catch-all results will also go through the matching filtering and diversity processing logic in S2. If the search results are still empty after removing some unsuitable jobs through matching filtering, the product interaction will be guided to catch-all results (such as redirecting to a job landing page with a larger job pool), and then the results will be returned to the user.

[0025] In summary, this invention establishes a fallback keyword library. This library provides fallback keywords for blank tag pages when the matching of job postings in the tag pool with user requests fails. These blank tag pages can then display results based on these fallback keywords, thereby improving user experience and ensuring exposure for certain job postings. This effectively utilizes traffic and guarantees the effectiveness of the part-time job recommendation system. The use of fallback keywords ensures a certain relevance between user requests and returned job results, avoiding serious mismatches. The fallback keyword library employs a multi-level fallback approach, prioritizing the display of the most relevant keywords. The invention uses manual addition, allowing for the effective handling of extreme situations where synonyms associated with certain tag names cannot be matched in job postings. Manually increasing the weight of certain keywords further enhances the accuracy of the fallback. For unacceptable blank tag pages during deployment, this invention uses broad system keywords to minimize the blank rate, ensuring a good user experience while increasing traffic to relevant job postings and providing good exposure for businesses.

Claims

1. A part-time job recommendation system operation method with a bottom-up process, characterized by: The method comprises the following steps: S1, in the part-time job recommendation system, according to the actual business needs, the operation will the jobs meeting a certain characteristic be marked with the same label and placed in a label job pool to display in the job platform or put outside in the form of label landing page; S2, when the user requests, the label job pool is matched, filtered and screened with the user request, the remaining jobs meeting the requirements are processed for precision and diversity after the matching, filtering and screening, and the results are returned to the user; If the remaining job list after the matching, filtering and screening is empty, it is considered that the label landing page is blank, and in the case of blank, the keyword extraction is performed on the part-time job description text information marked with the same label in step 1, the label name is formed, and then step S3 is entered; the precision and diversity processing refers to sorting the job list after the matching, filtering and screening according to the click rate model or the conversion rate model to form the precision job result, and when topN is taken, the precision job result is processed for diversity with the logic of interspersing and scattering the same brand and the same purpose; S3, the label name of the user request is input into the label bottom-up word library, the label association bottom-up keyword and weight are obtained in the label bottom-up word library according to the label name, the bottom-up keyword and weight are used to search the entire job pool, and the results are returned to the user; The label bottom-up word library is obtained by training a word2vec model using the text information in the part-time job library; then, the label name is used to obtain a plurality of similar words in the similar word model, and the similarity of each similar word to the label name is obtained, the similarity is used as the weight of keyword search matching, the similar words, the similarity and the weight are synchronized to the bottom-up keyword weight configuration table, and then the bottom-up keyword weight configuration table is imported into the Elasticsearch search dsl to form the label bottom-up word library; The label bottom-up word library manually adds a batch of bottom-up keywords and weight items for a specific label by artificial manual addition, and synchronizes the bottom-up keywords and weight items to the bottom-up keyword weight configuration table as the second level bottom-up; the label bottom-up word library also adds system big words and weight items for system level key aspects, and synchronizes the system big words and weight items to the bottom-up keyword weight configuration table as the bottommost level bottom-up. 2.The part-time job recommendation system operation method with the bottom-up process of claim 1, wherein: In step S2, the matching, filtering and screening include user basic information filtering, temporarily stopped recruitment job filtering, distance factor filtering and risk control filtering. 3.The part-time job recommendation system operation method with the bottom-up process of claim 1, wherein: In step 2, the keyword extraction is performed by the TextRank algorithm of NLP.

Images