Employee welfare demand prediction and dynamic recommendation method based on deep learning

Abstract

The invention discloses an employee welfare demand prediction and dynamic recommendation method based on deep learning, relates to the technical field of demand prediction recommendation, and is used for solving the problems that a traditional recommendation system often depends on static user preferences or single-dimensional historical data, most recommendation systems are based on a few factors and cannot adapt to changes of the user preferences in time, and the recommendation effect is poor. An existing recommendation algorithm usually depends on historical data of a single user to carry out recommendation, and collective behaviors and preferences of similar users are ignored. And more accurate personalized recommendation is realized by collecting wish item input of the employees and setting a recommended tag proportion in combination with a tag proportion. According to the feedback data and operation conditions of the employees, the system can dynamically adjust the proportion of the recommendation labels and the similarity threshold value, so that the recommendation effect is continuously optimized, different recommendation methods are selected according to different conditions, the recommendation flexibility and diversity are improved, and the diversified requirements of the employees can be better met.

Description

Technical Field

[0001] The present invention relates to the field of demand forecasting and recommendation technology, and more specifically, to a method for forecasting and dynamically recommending employee welfare demand based on deep learning. Background Art

[0002] This system encompasses multiple areas, from natural language processing, recommendation algorithms, and similarity calculation to feedback optimization and adaptive learning. By integrating these technologies, the system can provide employees with personalized, accurate item recommendations and continuously optimize itself through feedback and data updates. This recommendation system not only improves employee satisfaction but also enhances the system's intelligence.

[0003] The existing technology has the following deficiencies:

[0004] Traditional recommendation systems often rely on static user preferences or single-dimensional historical data, which can be incapable of adapting to changing user preferences. Most recommendation systems are based on a few factors, whereas an employee's wish list may contain a wide variety of items, tags, and needs, and these needs can be highly dynamic. By calculating similarity across multiple tags and adjusting the weight of tags that assimilate employee numbers, we can more comprehensively capture employees' personalized needs. Existing recommendation algorithms typically rely on the historical data of a single user, ignoring the collective behavior and preferences of similar users and failing to adjust recommendation strategies based on real-time feedback.

[0005] In view of the above problems, the present invention proposes a solution. Summary of the Invention

[0006] In order to overcome the above-mentioned defects of the prior art, an embodiment of the present invention provides an employee welfare demand prediction and dynamic recommendation method based on deep learning to solve the problems raised in the above-mentioned background technology.

[0007] To achieve the above object, the present invention provides the following technical solutions:

[0008] The method for predicting and dynamically recommending employee benefits needs based on deep learning includes the following steps:

[0009] Step S1: Call the employee list, send a wish item input instruction to each employee, receive the employee input results and merge them into a wish list, mark different employees' wish lists with different employee numbers, and set the recommended tag ratio based on the tag ratio of different wish items in each employee's wish list;

[0010] Step S2: Calculate the similarity between the desired item's recommended tags and the tags of items in the item library, modify the similarity threshold based on the number of desired item tags, screen multiple assimilated employee numbers based on the preset assimilation range and compare historical data, calculate the assimilated employee's wish ratio, and make comprehensive recommendations;

[0011] Step S3: Obtain recommendation feedback data, set a recommendation threshold, update the label ratio for employees who resend the desired item input instruction, calculate the assimilated wish ratio by calling the label ratio of the desired items in the assimilated employee sequence number, and update the employee's recommendation label ratio based on the recommendation feedback data;

[0012] Step S4: Monitor the employee's work status and update the label ratio frequency to select different modal recommendation methods.

[0013] In a preferred embodiment, step S1 includes the following contents:

[0014] Connect to the employee data source and extract basic information of all employees, obtain the employee list, send a wish item input instruction to each employee, collect all employees' wish item inputs, and merge them into an overall wish list. Mark the employee number corresponding to each wish item in the merged list;

[0015] Extract tags from multiple items in each employee's wish list, count the number of times each tag appears, and calculate its tag ratio. The tag ratio is calculated as the ratio of the number of times the tag appears to the total number of items in the employee's wish list. Record the tag ratio of each tag for each wish item.

[0016] In a preferred embodiment, step S2 includes the following:

[0017] Get the labels of the items in the item library and calculate the Jaccard similarity between the recommended labels of the desired items and the labels of the items in the item library. The formula is: Where set A is the set of recommended labels for desired items, set B is the set of labels for items in the item library, |A∩B| is the number of elements in the intersection of sets A and B, and |A∪B| is the number of elements in the union of sets A and B;

[0018] According to the calculated Jaccard similarity value, set an initial similarity threshold T init , filter out items with high similarity and select items with similarity greater than or equal to the threshold;

[0019] Analyze the number of tag types in the desired item recommendation tag set, and adjust the similarity threshold based on the number of tag types. Use the following formula to adjust the threshold: Where: T init is the initial threshold, k is the adjustment coefficient, NA is the number of label types, N max is the maximum number of tag types;

[0020] Get the historical data of employee wish list tags, and use the above Jaccard similarity algorithm to calculate the similarity between the historical data of each employee wish list tag and the tag of the current employee wish list. The formula is expressed as: Where set C is the set of historical data of each employee's wish list label, set D is the set of labels of the current employee's wish list, |C∩D| is the number of elements in the intersection of sets C and D, and |C∪D| is the number of elements in the union of sets C and D;

[0021] Set a threshold to screen assimilated employees. Set a similarity threshold T to screen employees with a high similarity to the current employee. When the Jaccard similarity value of an employee is greater than or equal to the threshold T, the employee is considered an "assimilated employee";

[0022] Calculate the assimilation desire ratio of recommended items among assimilated employees to obtain the recommendation weight of the item. Determine the recommendation frequency based on the weight and the similarity between the recommended label of the desired item and the label of the items in the item library. Combined with the label ratio of employees, generate the final recommended label.

[0023] In a preferred embodiment, step S3 includes the following contents:

[0024] Send a list of recommended items to employees and collect their feedback. The feedback data is scored on a scale of 1-5. The feedback data includes the following: acceptance score, purchase intention score, click frequency, and the maximum number of clicks for the recommended items. Combined with the item feedback scores, a recommendation threshold is set. If the threshold is exceeded, the item is considered a valid recommendation.

[0025] Calculate the weighted total score of each item, assigning different weights to each data according to the following rules: acceptance score weight: 0.3, purchase intention score weight: 0.5, click frequency weight: 0.2, and the item feedback score formula is expressed as:

[0026] Set a recommendation threshold for the total item feedback score to update the label ratio for employees who resend the desired item input instructions. Employees whose total item feedback score is lower than the threshold will resend the desired item input instructions, obtain the new label ratio, and combine it with the assimilated desire ratio of the assimilated employees to update the recommended label ratio of the employees. The formula is: recommended label ratio = w × new label ratio + (1-w) × assimilated desire ratio.

[0027] In a preferred embodiment, step S4 includes the following contents:

[0028] Monitor the degree of employee work completion, obtain employee participation and feedback frequency, and comprehensively judge the degree of employee work completion;

[0029] Select different modal recommendation methods based on employee work status and the frequency of label update, and decide whether to use a collaborative filtering-based recommendation method or a deep learning-based recommendation method:

[0030] After receiving the employee's work completion degree and the frequency of updating the label, the employee's work completion degree and the frequency of updating the label are defined as input variables and divided into different fuzzy sets respectively;

[0031] The modal recommendation method is taken as the output variable and divided into fuzzy sets;

[0032] Formulate fuzzy rules to describe the impact of employee completion and the frequency definition of updated label proportion on the modality recommendation method;

[0033] Fuzzy reasoning is performed based on fuzzy rules to determine the modal recommendation method.

[0034] The technical effects and advantages of the employee welfare demand prediction and dynamic recommendation method based on deep learning of the present invention are as follows:

[0035] By collecting employee desired item input and setting the recommended tag ratio based on the tag percentage, more accurate personalized recommendations are achieved. Based on employee feedback and work performance, the system dynamically adjusts the recommended tag ratio and similarity threshold to continuously optimize recommendation effectiveness. By calculating the assimilated employee desire ratios, it is possible to identify groups of employees with similar needs, facilitating group-based recommendations and services. By selecting different recommendation methods based on different situations, this increases the flexibility and diversity of recommendations, better meeting the diverse needs of employees. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 This is a structural diagram of the employee welfare demand prediction and dynamic recommendation method based on deep learning in the present invention. DETAILED DESCRIPTION

[0037] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0038] Example 1

[0039] The present invention discloses a method for predicting and dynamically recommending employee welfare needs based on deep learning, comprising the steps of:

[0040] Step S1: Call the employee list, send a wish item input instruction to each employee, receive the employee input results and merge them into a wish list, mark different employees' wish lists with different employee numbers, and set the recommended tag ratio based on the tag ratio of different wish items in each employee's wish list;

[0041] Step S2: Calculate the similarity between the desired item's recommended tags and the tags of items in the item library, modify the similarity threshold based on the number of desired item tags, screen multiple assimilated employee numbers based on the preset assimilation range and compare historical data, calculate the assimilated employee's wish ratio, and make comprehensive recommendations;

[0042] Step S3: Obtain recommendation feedback data, set a recommendation threshold, update the label ratio for employees who resend the desired item input instruction, calculate the assimilated wish ratio by calling the label ratio of the desired items in the assimilated employee sequence number, and update the employee's recommendation label ratio based on the recommendation feedback data;

[0043] Step S4: Monitor the employee's work status and update the label ratio frequency to select different modal recommendation methods.

[0044] The specific implementation is as follows:

[0045] In step S1, the employee list is called, and a desired item input instruction is sent to each employee. The employee input results are received and merged into a wish list. Different employee wish lists are marked with different employee serial numbers. The recommended tag ratio is set according to the tag ratio of different desired items in each employee's wish list. The specific content includes:

[0046] Connect to the employee data source and extract basic information of all employees. Obtain the employee list to ensure that every employee is on the list. Send a wish item input instruction to each employee, generate personalized wish item input instructions, collect all employees' wish item inputs, and merge them into an overall wish list. Mark the employee serial number corresponding to each wish item in the merged list to ensure that each item can be traced back to the corresponding employee.

[0047] To define the label ratios for different items in each employee's wish list, follow these steps:

[0048] Extract tags from the multiple items on each employee's wish list. These tags can be item type, brand, features, and more, depending on the data structure. For each employee's wish list, calculate the tag percentage by counting the number of times each tag appears. The tag percentage is calculated as the ratio of the number of times a tag appears to the total number of items on the employee's wish list. Record the tag percentage for each tag for each wish item.

[0049] In this embodiment, basic information of all employees is obtained from the data source for subsequent personalized processing, providing basic information for subsequent personalized tasks, ensuring the accuracy and comprehensiveness of the recommendation system, collecting each employee's specific desired items, improving employee participation and accuracy, facilitating subsequent data processing and recommendations, ensuring the traceability of desired items, and laying a solid foundation for further data analysis.

[0050] By tagging the employee information corresponding to each wish item, each wish item can be traced back to a specific employee, preventing data loss and helping maintain personalized recommendations and analysis. This allows you to trace back each employee's specific wishes for targeted analysis, quantify the characteristics of each employee's wish item, and analyze which characteristics are most important across employee groups, facilitating subsequent similarity analysis and recommendations. By analyzing historical data, other employees with similar wish items as a specific employee are included in the "assimilation scope," identifying similar employee groups. This provides data support for more precise personalized recommendations, uncovering groups of employees with similar needs, and further improving the accuracy of the recommendation system.

[0051] In step S2, the similarity between the recommended tags of the desired items and the tags of the items in the item library is calculated. The similarity threshold is modified based on the number of tags of the desired items. Based on the preset assimilation range and historical data, multiple assimilated employee numbers are screened, the wish ratio of assimilated employees is calculated, and comprehensive recommendations are made. The specific contents include:

[0052] Get the labels of the items in the item library, and calculate the similarity between the recommended labels of the desired items and the labels of the items in the item library. The formula is expressed as: Where set A is the set of recommended labels for desired items, set B is the set of labels for items in the item library, |A∩B| is the number of elements in the intersection of sets A and B, and |A∪B| is the number of elements in the union of sets A and B.

[0053] According to the calculated Jaccard similarity value, set an initial similarity threshold T init , filter out items with high similarity and only select items with similarity greater than or equal to the threshold. For example, if the initial threshold is set to 0.5, only those items with similarity higher than 0.5 will be recommended to employees.

[0054] It should be noted that Jaccard similarity is a method for measuring the similarity between two sets, and is widely used in fields such as information retrieval, text analysis, and bioinformatics. Its core idea is to measure the similarity between two sets by calculating the ratio of the intersection and the union of the two sets. The similarity value is between 0 and 1, where 0 means that the two sets have no identical elements and 1 means that the two sets are exactly the same. The larger the value, the higher the similarity between the two sets. An initial similarity threshold is set by professionals in this field based on experience or experimental data, which will not be elaborated here.

[0055] Analyze the number of tag types in the desired item recommendation tag set. If there are many tag types, it may mean that the recommendation needs are broader or more complex. In this case, you can appropriately lower the similarity threshold to include more potentially related items. If there are fewer tag types, it indicates that the needs are more specific. You can appropriately increase the similarity threshold to filter out more accurately matched items.

[0056] The specific rules for adjusting the threshold can be further refined based on experience, experimental data, or business needs. In this embodiment, the threshold is further refined based on the number of tag types to adjust the threshold using the following formula: Where: T init is the initial threshold, k is the adjustment coefficient, N A is the number of label types, N max is the maximum number of tag types. In this way, the recommendation system can dynamically adjust the screening threshold based on the breadth of user needs. Based on previous user feedback and behavior data, the adjustment coefficient k can be dynamically updated. For example, when the demand for certain tag types surges, k can be increased to allow the system to recommend a wider range of tag types, and vice versa.

[0057] Get the historical data of employee wish list tags, and use the above Jaccard similarity algorithm to calculate the similarity between the historical data of each employee wish list tag and the tag of the current employee wish list. The formula is expressed as: Where set C is the set of historical data of each employee's wish list label, set D is the set of labels of the current employee's wish list, |C∩D| is the number of elements in the intersection of sets C and D, and |C∪D| is the number of elements in the union of sets C and D.

[0058] Set a threshold to filter assimilated employees. Set a similarity threshold T to filter out employees with a high similarity to the current employee. For example, if the Jaccard similarity value of an employee is greater than or equal to the threshold T, the employee can be considered an "assimilated employee."

[0059] After obtaining assimilated employees, the system calculates the percentage of assimilated employees who would like to recommend a recommended item. By counting the percentage of assimilated employees who would like to choose a particular item, the system determines the item's recommendation weight. Based on these weighted weights and the similarity between the recommended item's label and the labels of items in the item library, the recommendation system determines the frequency of recommendations, specifically recommending more items preferred by assimilated employees. The system utilizes these employee label data to improve recommendation accuracy. The percentage of assimilated employees who would like to choose a particular item based on historical data is calculated and combined with the employee's label percentage to generate the final recommended label.

[0060] In step S3, the recommendation feedback data is obtained, and the recommendation threshold is set to update the label ratio of employees who resend the desired item input instruction. The label ratio of the desired items in the assimilated employee sequence number is called to calculate the assimilated wish ratio. The recommended label ratio of the employee is updated in combination with the recommendation feedback data. The specific content includes:

[0061] Send a list of recommended items to employees and collect their feedback, such as whether they are willing to accept the item and whether they have purchasing intentions. This feedback data can be collected through statistics and analysis, such as simple questionnaires, rating systems, or behavioral analysis. The feedback data is scored on a scale of 1-5, and each employee's feedback can be quantified using a rating system of 1-5. Feedback data includes the following: Acceptance score: whether the employee is willing to accept the item, on a scale of 1-5; Purchase intention score: whether the employee is willing to purchase the item, on a scale of 1-5; Click frequency: how many times employees click on each recommended item. Combined with the feedback preference value, a recommendation threshold is set. If the threshold is exceeded, the item is considered a valid recommendation.

[0062] Based on the above data, we calculated the weighted total score for each item, assigning different weights to each data item according to the following rules: acceptance score weight: 0.3, purchase intention score weight: 0.5, click frequency weight: 0.2. The item score formula is expressed as:

[0063] Set a recommended threshold for the total item feedback score to update the label ratio for employees who resend the desired item input instruction. For example, if the recommended threshold is set to 3, employees whose total item feedback score is lower than 3 need to resend the desired item input instruction, re-enter the desired item and update the label ratio.

[0064] For example, the feedback data collected from three employees is as follows:

[0065] staff Acceptance score Purchase intent score Click frequency Employee 1 4 3 20 Employee 2 2 1 50 Employee 3 5 4 10

[0066] Calculate the total item feedback score of the three employees as follows:

[0067] Employee 1's total item feedback score = (4 × 0.3) + (3 × 0.5) + (20 / 50 × 5 × 0.2) = 1.2 + 1.5 + 0.8 = 3.5;

[0068] Employee 2's total item feedback score = (2 × 0.3) + (1 × 0.5) + (50 / 50 × 5 × 0.2) = 0.6 + 0.5 + 1.0 = 2.1;

[0069] Employee 3's total item feedback score = (5 × 0.3) + (4 × 0.5) + (10 / 50 × 5 × 0.2) = 1.5 + 2.0 + 0.2 = 3.7;

[0070] Set the recommendation threshold to 3. Then: Employee 1 and Employee 3 have scores higher than 3 and are considered valid recommendations. Employee 2 has a score lower than 3 and needs to be re-evaluated.

[0071] Send a wish item input instruction to the employees who need to be re-evaluated, obtain the new label ratio and update the recommended label ratio of the employees in combination with the assimilated wish ratio of the assimilated employees. The formula is: Recommended label ratio = w × new label ratio + (1-w) × assimilated wish ratio, where: New label ratio is the new label ratio of the employees after re-evaluation, w is the weight coefficient, which indicates the degree of influence of the new label ratio on the final recommended label ratio, and the value range of the weight coefficient w is 0 to 1.

[0072] In step S4, the degree of completion of employee tasks and the frequency of label update are monitored to select different modal recommendation methods, including:

[0073] Monitor employee completion and obtain information on employee engagement and feedback frequency. Engagement: The degree to which employees proactively participate in their work. This can be measured by the number of tasks completed, the depth of their involvement, and the frequency of communication. This can be quantified by task completion rates and project participation. Feedback frequency: The frequency with which employees provide feedback on issues, suggestions, and work progress to superiors, colleagues, or management systems. This can be quantified by the number of monthly or weekly feedback sessions. Use this information to determine employee completion.

[0074] Select different modal recommendation methods based on employee work status and the frequency of label update, and decide whether to use a collaborative filtering-based recommendation method or a deep learning-based recommendation method:

[0075] After receiving the employee's job completion degree and the frequency of updated label proportion, the employee's job completion degree and the frequency of updated label proportion are defined as input variables and divided into different fuzzy sets respectively.

[0076] For example, "Low", "Medium", and "High" represent the degree of employee work completion, and "Low", "Medium", and "High" represent the frequency of update label proportions.

[0077] Define the modality recommendation method as the output variable and divide it into fuzzy sets, for example, "Deep Learning-based" and "Collaborative Filtering" for the modality recommendation method.

[0078] Formulate a set of fuzzy rules to describe the impact of different input variables on the output variable. The definition of rules can be based on professional knowledge or obtained through data analysis and experiments. For example:

[0079] Mark the employee's work completion degree as T, the frequency of updating the label as P, and the modal recommendation method as Method.

[0080] Rule 1:IF(T is High)AND(P is Low)THEN(Method is Col laborativeFiltering)

[0081] Rule 2:IF(T is Low)AND(P is High)THEN(Method is Deep Learning-based) ...

[0083] Fuzzy reasoning is performed based on fuzzy rules to determine the scheme of modal recommendation method.

[0084] The division of fuzzy sets can be adjusted according to actual conditions. For example, although this embodiment takes three fuzzy sets as an example, in fact, the degree of employee work completion, the frequency of updated label proportion and the modal recommendation method can be divided into more than three sets to facilitate better precise adjustment according to different temperatures.

[0085] It should be noted that the above formulas are all dimensionless and numerical calculations. The formula is a formula for the latest real situation obtained by collecting a large amount of data and performing software simulation. The preset parameters in the formula are set by technical personnel in this field according to actual conditions. Furthermore, for the judgment of the degree of employee completion and the frequency of updating label proportions, thresholds can be set according to actual conditions for judgment, which will not be elaborated here.

[0086] Collaborative filtering recommendation method: A method in traditional recommendation systems that makes recommendations based on the user's historical behavior data. This recommendation method is simple and easy to implement, does not require a complex model, does not rely on additional information about items or users, and makes recommendations based on the behavior data itself. However, if the user's behavior data is very sparse, the effectiveness of collaborative recommendation will drop significantly. For new users or new items, there is a lack of sufficient historical data, and the recommendation effect is poor. Collaborative recommendation methods can be generally divided into two categories: user-based collaborative recommendation and item-based collaborative recommendation:

[0087] User-based collaborative recommendation: If two users liked or rated similar items in the past, they are likely to like similar items in the future. By finding other users with similar behaviors to the target user, we can recommend items that these users like.

[0088] Item-based collaborative recommendation: If a user likes an item, they may like other items similar to it. By calculating the similarity between items, we can recommend other items similar to the items they already like to users.

[0089] Deep learning recommendation methods: A more complex recommendation method proposed with the development of deep learning technology. It can handle more complex data relationships, especially effectively learn nonlinear relationships between users and items. It can utilize a variety of data types, not just rating or click data. It is less affected by sparsity issues and can effectively learn features through deep network models. However, the training process is relatively complex and usually requires a large amount of computing resources and data support. The model is less interpretable, making it difficult to explain how the model makes recommendations. It requires a long training time, especially when the amount of data is large. Common recommendation system models include: matrix factorization: embedding user and item features into a low-dimensional space and optimizing them through neural networks; neural collaborative filtering: using neural networks to handle collaborative filtering problems and combining the advantages of deep learning; deep matrix factorization: combining traditional matrix factorization and deep neural network models, using multi-layer networks to learn potential feature representations.

[0090] The above formulas are all dimensionless and numerical calculations. The formulas are obtained by collecting a large amount of data and performing software simulation to obtain the most recent real situation. The preset parameters in the formulas are set by technicians in this field according to actual conditions.

[0091] The above embodiments may be implemented in whole or in part through software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments may be implemented in whole or in part in the form of a computer program product.

[0092] Those skilled in the art will appreciate that the modules and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application of the technical solution and the invention constraints. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0093] In addition, each functional module in each embodiment of the present application may be integrated into one processing module, or each module may exist physically separately, or two or more modules may be integrated into one module.

[0094] The above description is merely a specific embodiment of the present application, but the scope of protection of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this application should be included in the scope of protection of this application. Therefore, the scope of protection of this application should be based on the scope of protection of the claims.

[0095] Finally: The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for predicting and dynamically recommending employee benefits needs based on deep learning, characterized by: The steps include: Step S1: Call the employee list, send a wish item input instruction to each employee, receive the employee input results and merge them into a wish list, mark different employees' wish lists with different employee numbers, and set the recommended tag ratio based on the tag ratio of different wish items in each employee's wish list; Step S2: Calculate the similarity between the desired item's recommended tags and the tags of items in the item library, modify the similarity threshold based on the number of desired item tags, screen multiple assimilated employee numbers based on the preset assimilation range and compare historical data, calculate the assimilated employee's wish ratio, and make comprehensive recommendations; Step S3: Obtain recommendation feedback data, set a recommendation threshold, update the label ratio for employees who resend the desired item input instruction, calculate the assimilated wish ratio by calling the label ratio of the desired items in the assimilated employee sequence number, and update the employee's recommendation label ratio based on the recommendation feedback data; Step S4: Monitor the employee's work status and update the label ratio frequency to select different modal recommendation methods.

2. The method for predicting and dynamically recommending employee benefits based on deep learning according to claim 1, characterized in that: Connect to the employee data source, extract the basic information of all employees, obtain the employee list, send a wish item input instruction to each employee, collect all employees' wish item inputs, and merge them into an overall wish list. Mark the employee serial number corresponding to each wish item in the merged list.

3. The method for predicting and dynamically recommending employee benefits based on deep learning according to claim 2, characterized in that: Extract tags from multiple items in each employee's wish list, count the number of times each tag appears, and calculate its tag ratio. The tag ratio is calculated as the ratio of the number of times the tag appears to the total number of items in the employee's wish list. Record the tag ratio of each tag for each wish item. Get the labels of the items in the item library and calculate the Jaccard similarity between the recommended labels of the desired items and the labels of the items in the item library. The formula is: Where set A is the set of recommended labels for desired items, set B is the set of labels for items in the item library, |A∩B| is the number of elements in the intersection of sets A and B, and |A∪B| is the number of elements in the union of sets A and B; According to the calculated Jaccard similarity value, set an initial similarity threshold T init , filter out items with higher similarity and select items with similarity greater than or equal to the threshold.

4. The method for predicting and dynamically recommending employee welfare needs based on deep learning according to claim 3 is characterized in that ； Analyze the number of tag types in the desired item recommendation tag set, and adjust the similarity threshold based on the number of tag types. Use the following formula to adjust the threshold: Where: T init is the initial threshold, k is the adjustment coefficient, N A is the number of label types, N max The maximum number of tag types.

5. The method for predicting and dynamically recommending employee benefits based on deep learning according to claim 4 is characterized by: Get the historical data of employee wish list tags, and use the above Jaccard similarity algorithm to calculate the similarity between the historical data of each employee wish list tag and the tag of the current employee wish list. The formula is expressed as: Where set C is the set of historical data of each employee's wish list label, set D is the set of labels of the current employee's wish list, |C∩D| is the number of elements in the intersection of sets C and D, and |C∪D| is the number of elements in the union of sets C and D.

6. The method for predicting and dynamically recommending employee benefits based on deep learning according to claim 5, characterized in that: Set a threshold to filter assimilated employees. Set a similarity threshold T to filter employees with a high similarity to the current employee. When the Jaccard similarity value of an employee is greater than or equal to the threshold T, the employee is considered an "assimilated employee"; Calculate the assimilation desire ratio of recommended items among assimilated employees to obtain the recommendation weight of the item. Determine the recommendation frequency based on the weight and the similarity between the recommended label of the desired item and the label of the items in the item library. Combined with the label ratio of employees, generate the final recommended label.

7. The method for predicting and dynamically recommending employee benefits based on deep learning according to claim 6, characterized in that: Send a list of recommended items to employees and collect employee feedback. The feedback data is scored from 1 to 5. The feedback data includes the following content: acceptance score, purchase intention score, click frequency, and the maximum number of clicks on the recommended items. Combined with the item feedback score, a recommendation threshold is set. When the threshold is higher than this, the item is considered a valid recommendation.

8. The method for predicting and dynamically recommending employee benefits based on deep learning according to claim 7, characterized in that: Calculate the weighted total score of each item, assigning different weights to each data according to the following rules: acceptance score weight: 0.3, purchase intention score weight: 0.5, click frequency weight: 0.2, and the item feedback score formula is expressed as: Set a recommendation threshold for the total item feedback score to update the label ratio for employees who resend the desired item input instructions. Employees whose total item feedback score is lower than the threshold will resend the desired item input instructions, obtain the new label ratio, and combine it with the assimilated desire ratio of the assimilated employees to update the recommended label ratio of the employees. The formula is: recommended label ratio = w × new label ratio + (1-w) × assimilated desire ratio.

9. The method for predicting and dynamically recommending employee benefits based on deep learning according to claim 8, characterized in that: Monitor the degree of employee work completion, obtain employee participation and feedback frequency, and comprehensively judge the degree of employee work completion.

10. The method for predicting and dynamically recommending employee benefits based on deep learning according to claim 9, characterized in that: Select different modal recommendation methods based on employee work status and the frequency of label update, and decide whether to use a collaborative filtering-based recommendation method or a deep learning-based recommendation method: After receiving the employee's work completion degree and the frequency of updating the label, the employee's work completion degree and the frequency of updating the label are defined as input variables and divided into different fuzzy sets respectively; The modal recommendation method is taken as the output variable and divided into fuzzy sets; Formulate fuzzy rules to describe the impact of employee completion and the frequency definition of updated label proportion on the modality recommendation method; Fuzzy reasoning is performed based on fuzzy rules to determine the modal recommendation method.

Images