Method and system for intelligent document recommendation based on multi-dimensional evaluation and dynamic weight

Abstract

The application belongs to the field of document recommendation, and provides a document intelligent recommendation method and system fusing multi-dimensional evaluation and dynamic weight to solve the problem that the personalization and precision of the current recommendation result are not ideal. The document intelligent recommendation method fusing multi-dimensional evaluation and dynamic weight comprises the following steps: in the cold start stage, a three-layer decision model is used to recommend a document template; in the hot data stage, according to the interactive behavior data of the user and the document template, the user portrait features are dynamically updated, the weight of the recommendation influencing factor is adaptively and dynamically adjusted, a document template recommendation list is generated according to the weighted summation method of the recommendation influencing factor. The accurate recommendation problem of new users without behavior data can be effectively solved; meanwhile, a dynamic weight adjustment mechanism is designed to realize the smooth transition from the cold start stage to the hot data stage, break through the bottleneck that the traditional recommendation system cannot adapt to the dynamic change of user behavior, and can process multiple types of professional documents.

Description

TECHNICAL FIELD

[0001] The present application belongs to the field of document recommendation, and particularly relates to a document intelligent recommendation method and system fusing multi-dimensional evaluation and dynamic weight. BACKGROUND

[0002] The statements in this section merely provide background information related to the present application and do not necessarily constitute the prior art.

[0003] For the professional document (such as contract template, legal instrument, etc.) recommendation field, the existing recommendation system usually adopts natural language processing (NLP) technology to perform semantic analysis on the document content, and realizes the matching between the user and the document through the collaborative filtering algorithm. The existing recommendation algorithm performs poorly when dealing with the cold start problem. When facing a new user, due to the lack of sufficient behavior data support, the system can only provide random recommendation or recommendation based on the popularity, and cannot objectively evaluate the quality of the professional document, so that the new document or the new user is difficult to obtain effective recommendation. In addition, when dealing with multi-source data, the existing system is difficult to fully integrate the multi-dimensional factors such as document quality, user industry characteristics, and use frequency, resulting in that the individualization and precision of the recommendation result are not ideal. SUMMARY

[0004] In order to solve the above technical problems, the present application provides a document intelligent recommendation method and system fusing multi-dimensional evaluation and dynamic weight, which can effectively solve the accurate recommendation problem of new users without behavior data; at the same time, a dynamic weight adjustment mechanism is designed to realize the smooth transition from the cold start to the hot data stage, and the system can automatically optimize the weight distribution of each factor according to the user behavior data, so that the recommendation result evolves from universal recommendation to individualized and accurate recommendation, breaks through the bottleneck that the traditional recommendation system cannot adapt to the dynamic change of user behavior, and can process various professional documents.

[0005] In order to achieve the above purpose, the present application adopts the following technical scheme:

[0006] The first aspect of the present application provides a document intelligent recommendation method fusing multi-dimensional evaluation and dynamic weight.

[0007] In one or more embodiments, a document intelligent recommendation method fusing multi-dimensional evaluation and dynamic weight is provided, comprising:

[0008] In the cold start stage, a three-layer decision model is used to recommend the document template; the three-layer decision model includes a target layer, a criterion layer and an index layer; the target layer takes the optimal selection of the document template recommendation as the decision target; the criterion layer includes three evaluation dimensions of quality score, industry matching degree and random exploration factor; the index layer includes the indexes of each evaluation dimension of the criterion layer;

[0009] In the hot data stage, the user portrait feature is dynamically updated according to the interaction behavior data of the user and the document template, so as to adaptively and dynamically adjust the weight of the recommendation influencing factor, and generate a document template recommendation list according to the weighted summation method of the recommendation influencing factor.

[0010] As an implementation mode, the user portrait feature includes an industry feature, a preference feature and a user behavior mode.

[0011] As an implementation mode, the recommendation influencing factor includes a quality score, an industry matching degree, a usage frequency and collaborative filtering.

[0012] As an implementation mode, in the process of adaptively and dynamically adjusting the weight of the recommendation influencing factor, a minimum weight threshold of the quality score is set, and the amount of the interaction behavior data of the user and the document template and the weight of the quality score are negatively correlated.

[0013] As an implementation mode, in the process of adaptively and dynamically adjusting the weight of the recommendation influencing factor, the weight of the industry matching degree is positively correlated with the industry concentration.

[0014] As an implementation mode, in the process of adaptively and dynamically adjusting the weight of the recommendation influencing factor, the weight of the usage frequency is represented by a segmented function, when the usage frequency is lower than or equal to a set threshold, the weight of the usage frequency adopts a preset basic weight, and when the usage frequency is higher than the set threshold, the weight of the usage frequency is positively correlated with the total amount of interaction behavior data.

[0015] As an implementation mode, in the process of adaptively and dynamically adjusting the weight of the recommendation influencing factor, the weight of the collaborative filtering is positively correlated with the size of the similar user group.

[0016] As an implementation mode, the weight of the recommendation influencing factor is dynamically optimized based on online learning, which is expressed as:

[0017] ;

[0018] wherein, is a learning rate; the function is constructed based on a user satisfaction index; represents the weight of the recommendation influencing factor before updating; represents the weight of the recommendation influencing factor after updating.

[0019] The second aspect of the application provides a document intelligent recommendation system integrating multi-dimensional evaluation and dynamic weight.

[0020] In one or more embodiments, a document intelligent recommendation system integrating multi-dimensional evaluation and dynamic weight comprises:

[0021] The cold start recommendation module is used in a cold start stage, and utilizes a three-layer decision model to recommend a document template; the three-layer decision model comprises a target layer, a criterion layer and an index layer; the target layer takes optimal selection of the document template recommendation as a decision target; the criterion layer comprises three evaluation dimensions of quality score, industry matching degree and random exploration factor; and the index layer comprises indexes of each evaluation dimension of the criterion layer;

[0022] The hot data recommendation module is used in a hot data stage, dynamically updates user portrait features according to user interaction behavior data with the document template, adaptively and dynamically adjusts weights of recommendation influence factors, generates a document template recommendation list according to a recommendation influence factor weighted summation method.

[0023] The third aspect of the present application provides an electronic device.

[0024] An electronic device comprises a memory, a processor, and a computer program stored on the memory and executable on the processor, and the processor implements the steps of the document intelligent recommendation method with fused multi-dimensional evaluation and dynamic weight as described above when executing the program.

[0025] Compared with the prior art, the present application has the following beneficial effects:

[0026] In the cold start stage, the present application adopts a three-layer decision model to recommend a document template, takes quality score as a dominant basis, takes industry matching degree as an auxiliary reference factor, and reserves necessary random exploration space, which not only ensures preferential recommendation of high-quality contract templates, but also reserves a small amount of random exploration space to avoid the recommended result from falling into local optimization, thereby ensuring that the recommended result has both professionalism and diversity.

[0027] In the hot data stage, the present application dynamically updates user portrait features according to user interaction behavior data with the document template, adaptively and dynamically adjusts weights of recommendation influence factors, generates a document template recommendation list according to a recommendation influence factor weighted summation method, so that the recommendation system realizes personalized and accurate recommendation when facing different users and different scenarios. BRIEF DESCRIPTION OF DRAWINGS

[0028] The drawings accompanying the specification of the present application form part of the present application and serve to further understand the present application, the illustrative embodiments of the present application and the description thereof serve to explain the present application and do not constitute an improper limitation of the present application.

[0029] Figure 1 is a flowchart of the document intelligent recommendation method with fused multi-dimensional evaluation and dynamic weight of the embodiment of the present application;

[0030] Figure 2 is a structural schematic diagram of the document intelligent recommendation system with fused multi-dimensional evaluation and dynamic weight of the embodiment of the present application;

[0031] Figure 3 This is a schematic diagram of an electronic device according to an embodiment of the present invention. Detailed Implementation

[0032] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0033] It should be noted that the following detailed description is illustrative and intended to provide further explanation of the invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0034] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the scope of exemplary embodiments according to the invention. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0035] Figure 1 This is a flowchart illustrating a document intelligent recommendation method that integrates multi-dimensional evaluation and dynamic weights according to an embodiment of the present invention. Figure 1 The document intelligent recommendation method that integrates multi-dimensional evaluation and dynamic weights in this embodiment may include the following steps S101~S102.

[0036] The specific implementation process of steps S101 to S102 is as follows:

[0037] Step S101: In the cold start phase, a three-layer decision model is used to recommend document templates; the three-layer decision model includes a target layer, a criterion layer, and an indicator layer; the target layer takes recommending the optimal choice of document template as the decision objective; the criterion layer includes three evaluation dimensions: quality score, industry matching degree, and random exploration factor; the indicator layer includes indicators for each evaluation dimension of the criterion layer.

[0038] In the embodiments of the present application, the three-layer decision model is constructed based on the analytic hierarchy process (AHP). In view of the uncertainty of industry feature recognition caused by the lack of user behavior data in the cold start stage, the conservative weight distribution strategy is adopted in the embodiments of the present application, so that the quality score is used as the dominant basis, the industry matching degree is used as the auxiliary reference factor, and the necessary random exploration space is reserved. This weight distribution scheme not only ensures the priority recommendation of high-quality contract templates, but also reserves a small amount of random exploration space to avoid the recommended results falling into local optimization, so as to ensure that the recommended results have both professionalism and diversity. The system uses the eigenvector method to calculate the judgment matrix of each level, solves the maximum eigenvalue and the corresponding eigenvector, and obtains the final weight of each evaluation factor after normalization. The numerical stable eigenvalue solving algorithm is used in the calculation process, and the convergence threshold is set to ensure the calculation accuracy of the weight distribution result.

[0039] In the design of the index layer, the quality score dimension is further refined into five sub-indices of clause integrity, compliance, structure specification, balance of rights and obligations, and risk control completeness, and the industry matching degree dimension includes two sub-indices of keyword coverage rate and professional term matching degree. The weights of each sub-index are pre-set by domain experts, forming a systematic and quantifiable evaluation framework, which effectively realizes the accurate matching of legal professional requirements and actual user needs.

[0040] The following will take the document template as an example to explain in detail:

[0041] In the quality score dimension, all templates in the contract template library are traversed, and the quality score is calculated from the five dimensions of clause integrity, compliance, structure specification, balance of rights and obligations, and risk control.

[0042] Clause completeness is based on legal knowledge graph comparison of essential clauses coverage (such as a sales contract needs to include basic clauses such as subject matter, price, and delivery method); compliance is based on rule engine matching mandatory regulations and identifying high-risk expressions through large language models; structure specification is based on clearness of clause serial number level and consistency of format to improve contract readability; balance of rights and obligations is based on large model evaluation of rationality of division of rights and obligations of both parties; risk control completeness is judged by detecting risk control clauses such as default clauses, dispute resolution mechanisms, and confidentiality obligations and evaluating the rationality of liability limitation clauses. In the quality scoring model, dynamic weights are assigned to the five evaluation dimensions according to the nature of the document: the weights are dynamically adjusted according to industry attributes, document complexity, and user roles, so that the quality evaluation model can automatically adapt to different scene needs. In the cold start stage, the recommended results are quickly generated by presetting industry benchmark weights, and the personalized weight allocation is gradually optimized as the user behavior data accumulates. For example, a sales contract may focus more on clause completeness and risk control, a labor contract may focus more on compliance and balance of rights and obligations, and a lease contract may focus more on structure specification and compliance, and accordingly a larger weight is allocated.

[0043] In the industry matching degree evaluation dimension, the user's industry characteristics are extracted, and key word matching is performed through pre-constructed legal terminology dictionaries for finance, manufacturing, IT, etc. First, the TF-IDF algorithm is used to calculate the coverage of the template and the user's industry keywords, then a small model (BERT, etc.) is used to calculate the semantic similarity of the template terms and the industry terms, and finally the two indicators are weighted and summed according to the preset weight ratio to obtain an industry matching degree score in the range of 0-1. This way, preliminary personalized recommendation is achieved without user behavior data.

[0044] In the random exploration factor evaluation dimension, a random exploration mechanism is implemented according to a certain weight ratio (assuming 10%), and an ε-greedy strategy is used to randomly select candidate templates from the high-quality contract template pool.

[0045] The specific execution process is as follows: a random number between 0 and 1 is generated, if the value is less than 0.1, a template is randomly selected from the high-quality pool, otherwise the template is selected according to the weighted score. Step 3 avoids over-concentration of recommended results in a small number of high-score templates, ensuring the diversity of recommendations.

[0046] The quality score ( , corresponding weight ), industry matching degree ( , corresponding weight ), and random exploration factor ( , corresponding weight ) are linearly weighted and summed, and the recommended score calculation formula is: .

[0047] All candidate templates are ranked in descending order of scores, and top-N recommended results are selected and presented to users. The embodiment of the application integrates multi-dimensional evaluation factors to realize high-quality recommendation in the cold start stage.

[0048] In some optional embodiments, a contract template pre-classification index is also pre-constructed, high-quality templates are pre-clustered according to industries, and real-time calculation overhead is reduced; a cache mechanism is introduced to perform 30-minute short-term caching on recommended results of high-frequency access industries (such as manufacturing industry). In this way, the recommendation response speed in the cold start stage is significantly improved, and the system performance is guaranteed.

[0049] Step S102: In the hot data stage, the user portrait features are dynamically updated according to the user interaction behavior data with the document templates, so as to adaptively and dynamically adjust the weights of the recommendation influencing factors, and generate a document template recommendation list according to the weighted summation method of the recommendation influencing factors.

[0050] With the accumulation of user behavior data, the weight proportions of the quality score, industry relevance, usage frequency and collaborative filtering and other factors are automatically adjusted according to the richness of the user behavior data, so as to realize smooth transition from the cold start stage dominated by quality evaluation to the hot data stage of multi-factor collaborative optimization.

[0051] The user interaction behavior data with the document templates are continuously tracked and recorded, the interaction behavior data are divided into explicit behavior data (clicking and browsing, downloading, collecting, actual use, etc.) and implicit behavior data (page browsing time, number of times of viewing the same template), the behavior weight mapping function is designed according to the user behavior, different types of behavior are mapped into basic scores, and the higher the importance of the behavior is, the greater the basic score is. Meanwhile, a time decay coefficient The historical behavior value is corrected according to the time interval between the behavior occurrence time and the current time The more remote the behavior is, the more significant the decay of the value is, and the behavior score is shows an exponential decline trend with time: , wherein, is the initial setting of the user behavior basic score.

[0052] The user portrait features are extracted from the user interaction behavior data with the document templates.

[0053] The user portrait features include industry features, preference features and user behavior patterns.

[0054] The industry features: the industry distribution of the contract templates frequently used by the user is counted, and an industry concentration score is calculated;

[0055] Preference characteristics: the user's download template clause characteristics can be analyzed by the existing large language model to construct personalized clause preference labels, accurately capturing the user's special needs in clause content (such as "high-risk clause tolerance" and "dispute resolution mechanism inclination");

[0056] User behavior patterns: analyze user behavior sequences through existing clustering algorithms to identify different types of user behavior patterns (such as high and low operation frequency, operation interval length, etc., classified as "urgent" or "cautious").

[0057] In the embodiments of the present application, the recommendation influencing factors include quality score, industry matching degree, usage frequency and collaborative filtering.

[0058] In the process of adaptively and dynamically adjusting the weights of the recommendation influencing factors, a minimum weight threshold of the quality score is set, and the amount of user-document interaction behavior data is negatively related to the weight of the quality score.

[0059] For example, the weight of the quality score is ;

[0060] wherein, is the amount of user-document interaction behavior data (such as the number of clicks, browsing time, etc.); is a preset minimum weight threshold (such as 0.3); is an adjustment coefficient used to control the decay rate ( ); is a smoothing constant (to prevent weight explosion, usually taking 1).

[0061] It should be noted that the minimum weight threshold of the quality score can be set according to actual conditions. The amount of user-document interaction behavior data is negatively related to the weight of the quality score, which can be calculated according to actual conditions to determine the specific negative correlation function.

[0062] Specifically, in the process of adaptively and dynamically adjusting the weights of the recommendation influencing factors, the weight of the industry matching degree is positively related to the industry concentration.

[0063] For example, the weight of the industry matching degree is ; ;

[0064] wherein, is a positive correlation coefficient ( ), used to control the weight range (assuming it is set to 0.5, the weight range is mapped to [0, 0.5]); is the industry concentration (the value range is generally [0, 1]), which can be calculated by HHI or entropy index. represents the The proportion of a sub-industry in the user behavior or document set; The total number of industries. Industry concentration The higher (users or documents focus on a few industries), the greater the weight of the industry matching degree .

[0065] The industry matching degree here can be represented according to the similarity between the industry to which the document template belongs and the industry in the user interaction behavior data. The weight of the industry matching degree and the industry concentration are positively correlated, which can be set according to actual conditions.

[0066] In the process of adaptively and dynamically adjusting the weight of the recommendation influencing factor, the weight of the use frequency is represented by a segmented function. When the use frequency is less than or equal to a set threshold, the weight of the use frequency adopts a preset basic weight; when the use frequency is higher than the set threshold, the weight of the use frequency is positively correlated with the total amount of interaction behavior data.

[0067] ;

[0068] wherein, is the weight of the use frequency, is the use frequency (such as the number of clicks, the number of visits, etc.) of the current user or document, is a preset threshold value to distinguish between low-frequency and high-frequency behaviors, is the basic weight (fixed weight in the low-frequency stage), is the total amount of user-document interaction behavior data (such as total clicks, total use times, etc.), is the adjustment coefficient (controls the amplitude of the weight growth with the data amount, which needs to satisfy ).

[0069] The preset basic weight here can be set according to actual conditions. The weight of the use frequency and the total amount of interaction behavior data are positively correlated, which can be calculated to determine the specific positive correlation function according to actual conditions.

[0070] In the process of adaptively and dynamically adjusting the weight of the recommendation influencing factor, the weight of the collaborative filtering is positively correlated with the size of the similar user group.

[0071] ;

[0072] wherein, is the weight of the collaborative filtering; represents the size of the similar user group (i.e. the number of users similar to the target user behavior); is the maximum weight upper limit (controls the weight range, assuming it is set to 0.6); For smoothing coefficient, used to control the growth curve shape, for example, the value range is usually [5, 20].

[0073] The size of the similar user group here can be represented by the number of similar users, and the weight of collaborative filtering is positively related to the size of the similar user group. The specific positive correlation function can be calculated and determined according to the actual situation.

[0074] The final recommendation list is generated by multi-factor weighted fusion calculation, considering factors such as quality score, industry relevance, usage frequency and collaborative filtering. The weight of each factor is adjusted in real time according to the dynamic weight calculation model to adapt to the evolution of user preferences, thereby generating a recommendation score:

[0075] , wherein, , , , are the weights of quality score, industry relevance, usage frequency score and collaborative filtering score, , , , are the quality score, industry relevance, usage frequency score and collaborative filtering score.

[0076] On this basis, in order to improve the recommendation effect, a composite optimization mechanism is adopted:

[0077] First, introduce a random disturbance term to fine-tune the recommendation score , and the score adjustment is , so as to avoid the recommendation result from being too concentrated on high-frequency templates and ensure the diversity of the recommendation;

[0078] Secondly, set a time decay compensation factor for the newly stored documents, which acts on the recommendation score, and the compensated score is: , wherein, is the storage time of the document, and the decay speed is adjusted by the constant coefficient to enhance the exposure opportunity of new templates. At the same time, the time decay compensation factor can guarantee the original recommendation score ranking. By adjusting the constant coefficient and the constant coefficient , the content liquidity and the recommendation stability can be balanced.

[0079] Finally, set a forced filtering rule to automatically remove high-risk document templates with a risk score exceeding a preset threshold: , wherein, is an indicator function, which is 1 when , otherwise 0. is a risk threshold. In this way, the reliability of the recommendation is ensured from three dimensions of score optimization, novelty guarantee and legal compliance.

[0080] In some embodiments, the weights of the recommendation influencing factors are dynamically optimized based on online learning, which is expressed as:

[0081]

[0082] wherein, is a learning rate; the function is constructed based on the user satisfaction index; represents the weight of the recommendation influencing factor before updating; represents the weight of the recommendation influencing factor after updating.

[0083] The embodiment of the present application collects behavior feedback data such as user click conversion rate and use duration in real time, constructs a loss function in combination with a preset satisfaction evaluation index, and adjusts the weight parameters of dimensions such as quality score, industry relevance, use frequency and collaborative filtering by using optimization algorithms such as gradient descent, so as to update the weight parameters of each dimension regularly.

[0084] As shown in Figure 2 The document intelligent recommendation system integrating multi-dimensional evaluation and dynamic weight provided by the embodiment of the present application can be implemented in a software manner, and the document intelligent recommendation system integrating multi-dimensional evaluation and dynamic weight comprises the following software modules: a cold start recommendation module 201 and a hot data recommendation module 202.

[0085] The functions of each software module in the document intelligent recommendation system integrating multi-dimensional evaluation and dynamic weight will be introduced below.

[0086] The cold start recommendation module 201 is used to recommend a document template in a cold start stage by using a three-layer decision model; the three-layer decision model comprises a target layer, a criterion layer and an index layer; the target layer takes the optimal selection of the document template recommendation as the decision target; the criterion layer comprises three evaluation dimensions of quality score, industry matching degree and random exploration factor; and the index layer comprises indexes of each evaluation dimension of the criterion layer.

[0087] The hot data recommendation module 202 is used to dynamically update user portrait features in a hot data stage according to user interaction behavior data with the document template, so as to adaptively and dynamically adjust the weights of the recommendation influencing factors, generate a document template recommendation list according to a weighted summation method of the recommendation influencing factors.

[0088] In the hot data recommendation module 202, the user portrait features comprise industry features, preference features and user behavior patterns.

[0089] ​The recommended influence factors include quality score, industry matching degree, usage frequency and collaborative filtering.

[0090] In the process of adaptively and dynamically adjusting the weights of the recommended influence factors, a minimum weight threshold of the quality score is set, and the amount of user interaction behavior data with the document template and the weight of the quality score are negatively correlated.

[0091] In the process of adaptively and dynamically adjusting the weights of the recommended influence factors, the weight of the industry matching degree is positively correlated with the industry concentration.

[0092] In the process of adaptively and dynamically adjusting the weights of the recommended influence factors, the weight of the usage frequency is represented by a segmented function, when the usage frequency is lower than or equal to a set threshold, the weight of the usage frequency adopts a preset basic weight, and when the usage frequency is higher than the set threshold, the weight of the usage frequency is positively correlated with the total amount of interaction behavior data.

[0093] In the process of adaptively and dynamically adjusting the weights of the recommended influence factors, the weight of the collaborative filtering is positively correlated with the size of the similar user group.

[0094] In some optional embodiments, the weights of the recommended influence factors are dynamically optimized based on online learning, which is expressed as:

[0095] ;

[0096] wherein, is a learning rate; the function is constructed based on a user satisfaction index; represents the weight of the recommended influence factor before updating; represents the weight of the recommended influence factor after updating.

[0097] It should be noted that each module in the document intelligent recommendation system of the embodiment of the present application which fuses multi-dimensional evaluation and dynamic weight corresponds to each step in the document intelligent recommendation method of the embodiment of the present application which fuses multi-dimensional evaluation and dynamic weight, and the specific implementation process is the same, which will not be repeated here.

[0098] The present application adopts a three-layer decision model to recommend document templates in the cold start stage, takes the quality score as the dominant basis, takes the industry matching degree as the auxiliary reference factor, and retains necessary random exploration space, which not only ensures the priority recommendation of high-quality contract templates, but also retains a small amount of random exploration space to avoid the recommended results falling into local optimum, so as to ensure that the recommended results have both professionalism and diversity.

[0099] In the hot data stage, the user portrait features are dynamically updated according to the user interaction behavior data with the document template, the weight of the recommendation influencing factor is adaptively and dynamically adjusted, the document template recommendation list is generated according to the weighted summation method of the recommendation influencing factor, so that the recommendation system realizes personalized and accurate recommendation when facing different users and different scenes.

[0100] The structure of the electronic device of the embodiment of the present application is described in detail below, Figure 3 The schematic diagram of the component structure of the electronic device provided by the embodiment of the present application can be understood, Figure 3 Only the exemplary structure of the electronic device is shown, not all structures, and the shown part or all structures can be implemented as needed.

[0101] The electronic device provided by the embodiment of the present application includes at least one processor 301, a memory 302, a user interface 303 and at least one network interface 304. Each component in the document intelligent recommendation system fusing multi-dimensional evaluation and dynamic weight is coupled together through a bus system 305. It can be understood that the bus system 305 is used to realize the connection communication between the components. In addition to including a data bus, the bus system 305 also includes a power bus, a control bus and a state signal bus. However, in order to clearly illustrate, all kinds of buses are marked as bus system 305 in Figure 3 .

[0102] The user interface 303 can include a display, a keyboard, a mouse, a trackball, a click wheel, a key, a button, a touchpad or a touch screen, etc.

[0103] It can be understood that the memory 302 can be a volatile memory or a non-volatile memory, and can also include both volatile and non-volatile memories. The memory 302 in the embodiment of the present application can store data to support the operation of the terminal. Examples of these data include any computer programs for operating on the terminal, such as operating systems and application programs. The operating system contains various system programs, such as framework layer, core library layer, driver layer, etc., for implementing various basic services and processing hardware-based tasks. The application program can include various application programs.

[0104] In some embodiments, the fusion multi-dimensional evaluation and dynamic weight document intelligent recommendation system provided by the embodiments of the present application can be implemented in a combination of software and hardware. For example, the fusion multi-dimensional evaluation and dynamic weight document intelligent recommendation system provided by the embodiments of the present application can be a hardware decoding processor form of processor programmed to execute the fusion multi-dimensional evaluation and dynamic weight document intelligent recommendation method provided by the embodiments of the present application. For example, the hardware decoding processor form of processor can be one or more application specific integrated circuits (ASICs), DSPs, programmable logic devices (PLDs), complex programmable logic devices (CPLDs), field programmable gate arrays (FPGAs), or other electronic elements.

[0105] For example, the processor 301 can be an integrated circuit chip with a processing capability of signals, such as a general-purpose processor, a digital signal processor (DSP), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc., where the general-purpose processor can be a microprocessor or any conventional processor.

[0106] For example, the processor 301 can be an integrated circuit chip with a processing capability of signals, such as a general-purpose processor, a digital signal processor (DSP), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc., where the general-purpose processor can be a microprocessor or any conventional processor.

[0107] The memory 302 in the embodiments of the present application is used to store various types of data to support the operation of the fusion multi-dimensional evaluation and dynamic weight document intelligent recommendation system, or to store programs for executing the fusion multi-dimensional evaluation and dynamic weight document intelligent recommendation method provided by the embodiments of the present application. Figure 1program code of the method. Examples of such data include any executable code such as an executable instruction for operating on a document intelligent recommendation system that fuses multi-dimensional evaluation with dynamic weight, and a program that implements the document intelligent recommendation method that fuses multi-dimensional evaluation with dynamic weight according to an embodiment of the present application can be included in the executable instruction.

[0108] In particular, according to embodiments of the present application, the processes described above with reference to the flowcharts can be implemented as a computer software program. For example, embodiments of the present application include a computer program product that includes a computer program tangibly embodied on a computer readable medium, the computer program including instructions for execution by a processor to perform the various functions defined in the apparatus of the present application. Figure 1 program code of the method. In such embodiments, the computer program can be downloaded and installed from a network via the communication section, and / or installed from a removable medium. When the computer program is executed by the central processing unit, various functions defined in the apparatus of the present application are performed.

[0109] The present application is described with reference to the flowcharts and / or block diagrams of the methods, apparatus (system) and computer program products of embodiments of the present application. It should be understood that each flow and / or block in the flowcharts and / or block diagrams, and combinations of flows and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowcharts and / or block diagrams block or blocks. Figure 1 one or more flows and / or blocks Figure 1 Figure 1 an apparatus that implements the functions specified in the flowchart block or blocks.

[0110] The above merely provides the preferred embodiments of the present application and is not intended to limit the present application. For those skilled in the art, the present application can have various modifications and changes. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present application shall fall within the scope of the present application.

Claims

1. A document intelligent recommendation method integrating multi-dimensional evaluation and dynamic weights, characterized in that, include: During the cold start phase, a three-level decision model is used to recommend document templates; The three-layer decision-making model comprises an objective layer, a criterion layer, and an indicator layer. The objective layer takes recommending the optimal choice of document template as the decision objective. The criterion layer includes three evaluation dimensions: quality score, industry matching degree, and random exploration factor. The indicator layer includes indicators for each evaluation dimension of the criterion layer. A conservative weight allocation strategy is adopted, making the quality score the primary basis, the industry matching degree the auxiliary reference factor, and retaining the necessary random exploration space. During the hot data phase, user profile features are dynamically updated based on user interaction data with document templates to adaptively adjust the weights of recommendation influencing factors. A document template recommendation list is generated based on the weighted summation method of recommendation influencing factors. The formula for the weighted summation is: ,in, , , , These are the weights for quality score, industry matching score, usage frequency score, and collaborative filtering score, respectively. , , , These are quality score, industry matching degree, usage frequency score, and collaborative filtering score, respectively. As user behavior data accumulates, the weight ratios of quality scores, industry matching, usage frequency, and collaborative filtering factors are automatically adjusted based on the richness of the user behavior data, achieving a smooth transition from quality assessment-led in the cold start phase to multi-factor collaborative optimization in the hot data phase. In the process of adaptively and dynamically adjusting the weights of recommendation influencing factors, a minimum weight threshold for quality scores is set, and the amount of user-document interaction data is negatively correlated with the weight of quality scores; in the process of adaptively and dynamically adjusting the weights of recommendation influencing factors, the weight of industry matching degree is positively correlated with industry concentration. In the process of adaptively and dynamically adjusting the weights of recommendation influencing factors, the weight of usage frequency is represented by a piecewise function. When the usage frequency is lower than or equal to a set threshold, the weight of usage frequency adopts a preset base weight; when the usage frequency is higher than the set threshold, the weight of usage frequency is positively correlated with the total amount of interaction data. In the process of adaptively and dynamically adjusting the weights of recommendation influencing factors, the weights of collaborative filtering are positively correlated with the size of similar user groups; Recommended influencing factors include quality score, industry fit, frequency of use, and collaborative filtering.

2. The document intelligent recommendation method integrating multi-dimensional evaluation and dynamic weights as described in claim 1, characterized in that, The user profile features include industry characteristics, preference characteristics, and user behavior patterns.

3. The document intelligent recommendation method integrating multi-dimensional evaluation and dynamic weights as described in claim 1, characterized in that, The weights of the recommendation influencing factors are dynamically optimized based on online learning, and are expressed as follows: ； in, The learning rate; The function is built based on user satisfaction metrics; This indicates the weight of the factors influencing the recommendations before the update; This indicates the weight of the factors influencing the updated recommendations.

4. A document intelligent recommendation system integrating multi-dimensional evaluation and dynamic weighting, characterized in that, include: The cold start recommendation module is used to recommend document templates during the cold start phase using a three-layer decision model. The three-layer decision-making model comprises an objective layer, a criterion layer, and an indicator layer. The objective layer takes recommending the optimal choice of document template as the decision objective. The criterion layer includes three evaluation dimensions: quality score, industry matching degree, and random exploration factor. The indicator layer includes indicators for each evaluation dimension of the criterion layer. A conservative weight allocation strategy is adopted, making the quality score the primary basis, the industry matching degree the auxiliary reference factor, and retaining the necessary random exploration space. The hot data recommendation module is used to dynamically update user profile features based on user interaction data with document templates during the hot data phase, so as to adaptively and dynamically adjust the weight of recommendation influencing factors, and generate a document template recommendation list according to the weighted summation method of recommendation influencing factors. The formula for the weighted summation is: ,in, , , , These are the weights for quality score, industry matching score, usage frequency score, and collaborative filtering score, respectively. , , , These are quality score, industry matching degree, usage frequency score, and collaborative filtering score, respectively. As user behavior data accumulates, the weight ratios of quality scores, industry matching, usage frequency, and collaborative filtering factors are automatically adjusted based on the richness of the user behavior data, achieving a smooth transition from quality assessment-led in the cold start phase to multi-factor collaborative optimization in the hot data phase. In the process of adaptively and dynamically adjusting the weights of recommendation influencing factors, a minimum weight threshold for quality scores is set, and the amount of user-document interaction data is negatively correlated with the weight of quality scores; in the process of adaptively and dynamically adjusting the weights of recommendation influencing factors, the weight of industry matching degree is positively correlated with industry concentration. In the process of adaptively and dynamically adjusting the weights of recommendation influencing factors, the weight of usage frequency is represented by a piecewise function. When the usage frequency is lower than or equal to a set threshold, the weight of usage frequency adopts a preset base weight; when the usage frequency is higher than the set threshold, the weight of usage frequency is positively correlated with the total amount of interaction data. In the process of adaptively and dynamically adjusting the weights of recommendation influencing factors, the weights of collaborative filtering are positively correlated with the size of similar user groups; Recommended influencing factors include quality score, industry fit, frequency of use, and collaborative filtering.

5. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the steps in the document intelligent recommendation method that integrates multi-dimensional evaluation and dynamic weights as described in any one of claims 1-3.

Images