Intelligent recommendation method and system for home decoration designers

Abstract

The invention relates to a home decoration designer intelligent recommendation method and system, and the method comprises the steps: obtaining user demand information and historical behavior information, and extracting demand features and historical behavior features; obtaining a designer database and performing classification to form a multi-dimensional feature vector; inputting the demand features, the historical behavior features and the multi-dimensional feature vectors into a matching model to generate a matching degree matrix; generating a demand solution of the matching degree matrix based on the loss function and a solution algorithm; generating a recommendation list based on the generated demand solution; according to the method, the demand features and the historical behavior features are extracted, the matching degree matrix is generated by combining the multi-dimensional feature vector input matching model, and the demand solution is generated based on the loss function and the solution algorithm, so that the recommendation list is generated and output, and dynamic preference mining of home decoration design and designer recommendation combined with the multi-dimensional features are realized; the method has the effect of improving the practicability of the recommendation system and the user satisfaction.

Description

Technical Field

[0001] This application relates to the technical field of intelligent analysis, and in particular, to an intelligent recommendation method and system for home improvement designers. Background Art

[0002] As an important part of the transformation of the living environment, the demand for personalized services in home improvement design is increasing. With the popularization of the Internet home improvement platform, how to efficiently match users with designers has become a key issue in improving the service efficiency of the industry.

[0003] Existing home improvement designer recommendation systems usually use methods such as keyword matching, collaborative filtering, or content similarity calculation for recommendation. For example, they directly compare with designer tags based on static requirements such as the budget and style preferences filled in by users.

[0004] However, the above methods have obvious limitations: First, they often tend to ignore the dynamic preferences hidden in the user's historical behavior, resulting in a lack of personalization in the recommendation results. Second, most systems only consider single - dimensional matching (such as style or price), and fail to comprehensively evaluate the multi - dimensional characteristics of designers (such as service response speed, geographical coverage, quality of past cases, etc.), making the recommendation results one - sided and lacking practicality. Summary of the Invention

[0005] To solve the above - mentioned defects, this application provides an intelligent recommendation method and system for home improvement designers.

[0006] The first invention object of this application is achieved through the following technical solutions:

[0007] An intelligent recommendation method for home improvement designers, including the steps of:

[0008] Obtain user demand information and historical behavior information, and extract demand features and historical behavior features respectively;

[0009] Obtain a designer database, and classify the designer information in the designer database based on a pre - set classification algorithm to correspondingly form multi - dimensional feature vectors associated with the designer information;

[0010] Input the demand features, historical behavior features, and multi - dimensional feature vectors into the same pre - trained matching model to generate a matching degree matrix;

[0011] When receiving the matching degree matrix output by the matching model, generate a number of demand solutions based on a predefined loss function and the solution algorithm associated with the loss function;

[0012] Generate a recommendation list based on the generated number of demand solutions and send it to the user terminal that sent the user demand information.

[0013] By adopting the above technical solution, this application captures the user's demand information and historical behavior information, extracts the corresponding demand features and historical behavior features, combines with the multi-dimensional feature vectors formed by the classification algorithm in the designer database, inputs the demand features, historical behavior features and designer multi-dimensional feature vectors into a pre-trained matching model to generate a matching degree matrix, and generates and outputs a recommendation list based on the loss function and the solution algorithm to solve the demand of the matching degree matrix, realizing the dynamic preference mining of home improvement design and the accurate recommendation of designers combined with multi-dimensional features: capturing the implicit dynamic preferences of users through the extraction of historical behavior features, and solving the problem of insufficient personalization caused by ignoring user behavior data in the prior art; at the same time, through the collaborative matching of the designer multi-dimensional feature vectors and demand features, corresponding to solve the limitations of single-dimensional recommendation, making the recommendation results not only comprehensively consider the comprehensive capabilities of designers, but also fit the actual needs of users, and having the effect of improving the practicality of the recommendation system and user satisfaction.

[0014] In a preferred example, this application can be further configured as follows: the steps of obtaining the user's demand information and historical behavior information, and respectively extracting the demand features and historical behavior features include the steps of:

[0015] Identifying the identity information corresponding to the user's demand information, and obtaining the historical behavior information of this identity information;

[0016] Inputting the historical behavior information into a pre-trained feature extraction model, so that the feature extraction model extracts and outputs historical behavior features based on the historical behavior information.

[0017] By adopting the above technical solution, this application identifies the identity information corresponding to the user's demand information and obtains its historical behavior data, inputs the historical behavior information into a pre-trained feature extraction model for historical behavior feature extraction, and realizes the mining and transformation of user behavior data: using the pre-trained feature extraction model to process the historical behavior information, automatically identifying the rules in unstructured data such as user browsing records, collection preferences, and interaction feedback, reducing subjectivity and inefficiency, not only significantly improving the accuracy and consistency of feature extraction, but also dynamically adapting to the behavior pattern differences of different users, and providing a behavior feature input with representativeness for the subsequent matching model.

[0018] In a preferred example, this application can be further configured as follows: the feature extraction model includes a quantization layer, an association layer, an identification layer, and an output layer. The historical behavior information includes browsing behavior, collection behavior, consultation behavior, and negative feedback. The steps of inputting the historical behavior information into a pre-trained feature extraction model, so that the feature extraction model extracts and outputs historical behavior features based on the historical behavior information include the steps of:

[0019] The quantization layer quantizes the browsing behavior and the collection behavior, and outputs a basic feature vector;

[0020] The association layer captures the association behaviors among browsing behaviors, collection behaviors, consultation behaviors, and negative feedback, optimizes the basic feature vector based on the association behaviors, and outputs the optimized feature vector;

[0021] The recognition layer generates implicit behavior vectors and abnormal behavior vectors based on browsing behavior, collection behavior, consultation behavior, and negative feedback;

[0022] The output layer generates historical behavior features based on the optimized feature vector, implicit behavior vector, and abnormal behavior vector.

[0023] By adopting the above technical solution, the present application generates basic feature vectors by quantifying the features of browsing behavior and collection behavior through the quantization layer, and outputs optimized feature vectors by analyzing the associated behaviors among browsing behavior, collection behavior, consulting behavior and negative feedback through the association layer, and combines the implicit behavior vectors and abnormal behavior vectors generated by the recognition layer, and finally outputs the historical behavior features through the output layer to realize multi-dimensional modeling of user behavior data; adopts a hierarchical feature extraction architecture, ensures the numerical expression of basic behavior features through the quantization layer, uses the association layer to mine the potential connections between multiple types of user behaviors to enhance the feature representation capability, and uses the recognition layer to capture the implicit patterns and abnormal feedback in user preferences, and finally realizes the comprehensive optimization of features through the output layer, so that the generated historical behavior features can not only reflect the user's explicit behavior preferences, but also identify implicit needs and sensitive issues, and provide more comprehensive and accurate user portrait data for subsequent matching models.

[0024] In a preferred example, the present application can be further configured as follows: the identification layer generates implicit behavior vectors and abnormal behavior vectors based on browsing behavior, collection behavior, consultation behavior and negative feedback, including the following steps:

[0025] The recognition layer performs text analysis on browsing and collection behaviors, and extracts preferred style tags and urgency tags;

[0026] Calculate the preference strength of the extracted preference style labels and the urgency strength of the urgency labels;

[0027] An implicit behavior vector is generated based on the preference strength and the urgency strength.

[0028] By adopting the above technical solutions, the present application performs text analysis on browsing behavior and collection behavior through the recognition layer to extract preference style tags and urgency tags, and calculates the preference intensity and urgency intensity to generate an implicit behavior vector, so as to realize the mining of users' potential design requirements and decision-making psychology; adopts text analysis technology to analyze the semantic information in users' behavior data, accurately captures the degree of users' preference for a specific design style by quantifying the preference intensity of the preference style tags, and at the same time analyzes the time sensitivity of users' decisions through the urgency intensity of the urgency tags, so that the generated implicit behavior vector can reflect both users' aesthetic preferences and project progress requirements, provides a dual reference basis including explicit and implicit requirements for designer matching, effectively improves the understanding ability of the recommendation system for users' complex intentions, and meets users' multi-dimensional requirements.

[0029] In a preferred example, the present application can be further configured as follows: The steps of the recognition layer generating an implicit behavior vector and an abnormal behavior vector based on browsing behavior, collection behavior, consultation behavior, and negative feedback include the steps of:

[0030] Extract the contradictory behavior characteristics in browsing behavior, collection behavior, consultation behavior, and negative feedback based on pre-set contradiction detection rules;

[0031] Extract the abnormal frequency characteristics in browsing behavior, collection behavior, consultation behavior, and negative feedback based on pre-set frequency detection rules;

[0032] Generate an abnormal behavior vector based on the contradictory behavior characteristics and the abnormal frequency characteristics.

[0033] By adopting the above technical solutions, the present application extracts the contradictory behavior characteristics in browsing behavior, collection behavior, consultation behavior, and negative feedback through pre-set contradiction detection rules, and at the same time uses frequency detection rules to extract the abnormal frequency characteristics in browsing behavior, collection behavior, consultation behavior, and negative feedback, and finally fuses them to generate an abnormal behavior vector, realizing the recognition of potential abnormal patterns in users' behavior data; adopts a dual detection mechanism, identifies the inconsistencies in users' behavior logic through contradiction detection rules, and at the same time captures abnormal operation patterns through frequency detection rules, so that the generated abnormal behavior vector can accurately reflect the hesitation, contradiction, and special needs in users' decision-making process, provides an important behavior anomaly warning index for the recommendation system, and thus improves the reliability of the recommendation results and user satisfaction.

[0034] In a preferred example, the present application can be further configured as follows: The matching model includes an embedding layer, an interaction layer, and an output layer. The steps of inputting the demand characteristics, historical behavior characteristics, and multi-dimensional feature vectors into the same pre-trained matching model to generate a matching degree matrix include the steps of:

[0035] The embedding layer maps the requirement features, historical behavior features, and multi-dimensional feature vectors to the same embedding space to generate a user embedding vector and a designer embedding vector;

[0036] The interaction layer identifies the interaction correlation between the user embedding vector and the designer embedding vector to generate an interaction feature vector;

[0037] The output layer generates a matching degree matrix based on the interaction feature vector.

[0038] By adopting the above technical solution, in this application, the requirement features, historical behavior features, and multi-dimensional feature vectors of designers are uniformly mapped to the same embedding space through the embedding layer to generate a user embedding vector and a designer embedding vector. Then, the interaction layer analyzes the interaction correlation between the two to generate an interaction feature vector. Finally, the output layer outputs a matching degree matrix to achieve the deep semantic matching between user requirements and designer features; the embedding space mapping technology is used to solve the problem that it is difficult to directly compare multi-source heterogeneous features. By projecting the features on the user side and the designer side into a unified vector space, features in different dimensions such as style preferences and service capabilities are comparable; the interaction layer can identify deep matching relationships that are difficult to discover by traditional methods by capturing potential correlation patterns between user and designer features, such as the fit between user implicit requirements and designer special expertise; the finally output matching degree matrix not only contains the similarity of surface features, but also contains the potential synergy of the user-designer combination, making the recommendation results not only meet the explicit needs of users, but also explore the potential matching possibilities that have not been clearly expressed, with the effect of significantly improving the intelligence level and accuracy of the recommendation system.

[0039] In a preferred example of this application, it can be further configured as follows: the step of generating a number of demand solutions based on a predefined loss function and the solution algorithm associated with the loss function when receiving the matching degree matrix output by the matching model includes the steps of:

[0040] Match a preset corresponding matching error loss function and diversity constraint loss function based on the matching degree matrix;

[0041] Construct a multi-objective loss function based on the matched matching error loss function and diversity constraint loss function, and match the associated solution algorithm;

[0042] Generate a set of recommended solutions based on the multi-objective loss function;

[0043] Generate a number of demand solutions for the set of recommended solutions based on the matched solution algorithm.

[0044] By adopting the above technical solution, the present application triggers the joint optimization of the matching error loss function and the diversity constraint loss function through the matching degree matrix, constructs a multi-objective loss function and generates a set of recommended solutions, and finally outputs several demand solutions through the solution algorithm, realizing the intelligent balance between the accuracy and diversity of the recommendation results; adopting a multi-objective optimization framework, ensuring the fit between the recommendation results and the user's needs through the matching error loss function, and at the same time using the diversity constraint loss function to avoid the homogenization of the recommendation list, so that the generated set of recommended solutions can not only accurately match the user's core needs, but also cover different types of designer resources; based on the further optimization of the set of recommended solutions by the solution algorithm, the recommendation strategy can be dynamically adjusted according to the actual business needs, realizing the reasonable distribution of the recommendation results on the premise of ensuring the recommendation quality, effectively solving the problem that the traditional recommendation system is prone to fall into local optimum and lead to the simplification of the recommendation list, improving both the user's selection space and satisfaction, and promoting the balanced exposure of the designer resources on the platform.

[0045] The above second invention object of the present application is achieved by the following technical solutions:

[0046] An intelligent recommendation system for home improvement designers, comprising:

[0047] A feature extraction module, configured to obtain user demand information and historical behavior information, and extract demand features and historical behavior features respectively;

[0048] A classification module, configured to obtain a designer database, and classify the designer information in the designer database based on a preset classification algorithm to correspondingly form a multi-dimensional feature vector associated with the designer information;

[0049] An input module, configured to input the demand features, historical behavior features, and multi-dimensional feature vectors into the same pre-trained matching model to generate a matching degree matrix;

[0050] A demand solution generation module, configured to generate several demand solutions based on a predefined loss function and a solution algorithm associated with the loss function when receiving the matching degree matrix output by the matching model;

[0051] A recommendation list generation module, configured to generate a recommendation list based on the generated several demand solutions and send it to the user terminal that sends the user demand information.

[0052] By adopting the above technical solution, a feature extraction module is configured to obtain user requirement information and historical behavior information, and extract requirement features and historical behavior features respectively; a classification module is configured to obtain a designer database and classify the designer information in the designer database based on a preset classification algorithm to correspondingly form a multi-dimensional feature vector associated with the designer information; an input module is configured to input the requirement features, historical behavior features and multi-dimensional feature vector into the same pre-trained matching model to generate a matching degree matrix; a requirement solution generation module is configured to, when receiving the matching degree matrix output by the matching model, generate a plurality of requirement solutions based on a predefined loss function and a solution algorithm associated with the loss function; and a recommendation list generation module is configured to generate a recommendation list based on the generated plurality of requirement solutions and send the recommendation list to the user terminal that sends the user requirement information.

[0053] In summary, the present application includes at least one of the following beneficial technical effects:

[0054] 1. By obtaining user requirement information and historical behavior information and extracting corresponding requirement features and historical behavior features, combining with the multi-dimensional feature vector formed by processing the designer database through a classification algorithm, inputting the requirement features, historical behavior features and designer multi-dimensional feature vector into a pre-trained matching model to generate a matching degree matrix, and generating requirement solutions for the matching degree matrix based on a loss function and a solution algorithm, so as to generate and output a recommendation list, the present application realizes dynamic preference mining for home decoration design and precise recommendation of designers combined with multi-dimensional features: by extracting historical behavior features to capture the implicit dynamic preferences of users, it solves the problem of insufficient personalization in the prior art due to ignoring user behavior data; at the same time, through the collaborative matching of the designer multi-dimensional feature vector and the requirement features, it correspondingly solves the limitation of single-dimensional recommendation, so that the recommendation result not only comprehensively considers the comprehensive ability of the designer, but also can meet the actual needs of users, and has the effect of improving the practicality of the recommendation system and user satisfaction;

[0055] 2. This application triggers the joint optimization of the matching error loss function and the diversity constraint loss function through the matching degree matrix, constructs a multi-objective loss function and generates a set of recommended solutions. Finally, several demand solutions are output through the solution algorithm to achieve an intelligent balance between the accuracy and diversity of the recommendation results; adopts a multi-objective optimization framework, ensures the fit between the recommendation results and user needs through the matching error loss function, and at the same time uses the diversity constraint loss function to avoid the homogenization of the recommendation list, so that the generated set of recommended solutions can accurately match the core needs of users and cover different types of designer resources; based on the further optimization of the set of recommended solutions by the solution algorithm, the recommendation strategy can be dynamically adjusted according to the actual business needs, and a reasonable distribution of the recommendation results can be achieved on the premise of ensuring the recommendation quality, effectively solving the problem that the traditional recommendation system is prone to fall into local optimality and lead to a single recommendation list, which not only improves the user's choice space and satisfaction, but also promotes the balanced exposure of the platform's designer resources. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] Figure 1 is a flowchart of an embodiment of an intelligent recommendation method for home improvement designers in this application;

[0057] Figure 2 is an implementation flowchart of step S12 in an embodiment of an intelligent recommendation method for home improvement designers in this application;

[0058] Figure 3 is an implementation flowchart of step S123 in an embodiment of an intelligent recommendation method for home improvement designers in this application;

[0059] Figure 4 is another implementation flowchart of step S123 in an embodiment of an intelligent recommendation method for home improvement designers in this application;

[0060] Figure 5 is an implementation flowchart of step S30 in an embodiment of an intelligent recommendation method for home improvement designers in this application;

[0061] Figure 6 is an implementation flowchart of step S40 in an embodiment of an intelligent recommendation method for home improvement designers in this application. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0062] The following will Figures 1-6 further describe this application in detail.

[0063] In one embodiment, as Figure 1 shown, this application discloses an intelligent recommendation method for home improvement designers, which specifically includes the following steps:

[0064] S10: Obtain user demand information and historical behavior information, and extract demand features and historical behavior features respectively;

[0065] In this embodiment, the user demand information is the demand expressed by the user based on home improvement design, such as including budget, style preference (such as modern, Nordic, Chinese, etc.), house area, functional requirements (such as storage, daylighting optimization, etc.); the historical behavior information is the operation record behavior information of the user on the platform; the demand feature is the structured data extracted from the user demand information, such as budget range, style label, functional label, etc.; the historical behavior feature is the implicit preference mined from the user's historical behavior, such as style tendency, decision-making urgency, sensitivity to the service quality of designers, etc.;

[0066] Specifically, obtain the user's home improvement demand information (such as budget, style preference, etc.) and historical behavior information, and extract demand features (explicit demands) and historical behavior features (implicit preferences) respectively.

[0067] S20: Obtain the designer database, and classify the designer information in the designer database based on a preset classification algorithm to correspondingly form a multi-dimensional feature vector associated with the designer information;

[0068] In this embodiment, the designer database is a structured data set storing designer information, including the designer's personal profile, case works, service evaluations, response speed, geographical coverage, etc.; the multi-dimensional feature vector is a numerical representation formed after classifying and quantifying the designer information, covering multiple dimensions such as professional skills, service capabilities, and user evaluations;

[0069] Specifically, read the designer information (such as case works, service scores, response speed, etc.) from the designer database, and classify the designers using a classification algorithm (such as clustering, label classification), and convert the designer information into a multi-dimensional feature vector (such as design style score, service score, response speed, etc.).

[0070] S30: Input the demand features, historical behavior features, and multi-dimensional feature vectors into the same pre-trained matching model to generate a matching degree matrix;

[0071] In this embodiment, the matching model is a pre-trained deep learning model for calculating the matching degree between user features and designer features; the matching degree matrix is a matching score matrix output by the matching model, reflecting the fitting degree of different combinations;

[0072] Specifically, input the user's demand features, historical behavior features, and the designer's multi-dimensional feature vectors into the matching model, so that the matching model calculates the matching degree between the user and each designer, and outputs the corresponding matching degree matrix, that is, the user-designer score table.

[0073] S40: When receiving the matching degree matrix output by the matching model, generate a number of solutions to be solved based on a predefined loss function and the solution algorithm associated with the loss function.

[0074] In this embodiment, the loss function is a predefined optimization objective function for evaluating matching errors and recommendation diversity; the solution algorithm is a mathematical method (such as gradient descent, genetic algorithm, etc.) applied to optimize the loss function and generate recommendation solutions; the solutions to be solved are candidate recommendation schemes generated after optimization.

[0075] Specifically, adjust the matching degree matrix through the predefined loss function and the solution algorithm associated with the loss function to generate multiple solutions to be solved, that is, candidate recommendation schemes.

[0076] S50: Generate a recommendation list based on the generated number of solutions to be solved and send it to the user terminal that sent the user demand information.

[0077] In this embodiment, the recommendation list is the designer recommendation ranking list finally presented to the user terminal.

[0078] Specifically, sort or fuse the solutions to be solved to generate the final recommendation list, and send the recommendation list to the user terminal for it to select or further consult.

[0079] In one embodiment, step S10 includes the steps:

[0080] S11: Identify the identity information corresponding to the user demand information and obtain the historical behavior information of the identity information.

[0081] S12: Input the historical behavior information into the pre-trained feature extraction model, so that the feature extraction model extracts and outputs historical behavior features based on the historical behavior information.

[0082] In this embodiment, the identity information is the unique identifier of the user, such as user ID, mobile phone number, account, etc., used to associate the historical data of the user.

[0083] Specifically, identify the identity information corresponding to the user demand information and obtain its historical behavior data, input the historical behavior information into the pre-trained feature extraction model for historical behavior feature extraction, and realize the mining and transformation of user behavior data.

[0084] In one embodiment, the feature extraction model includes a quantization layer, an association layer, an identification layer, and an output layer. The historical behavior information includes browsing behavior, collection behavior, consultation behavior, and negative feedback, such as Figure 2 As shown, step S12 includes the steps:

[0085] S121: The quantization layer quantizes the browsing behavior and the collection behavior and outputs a basic feature vector.

[0086] S122: The association layer captures the association behaviors among browsing behaviors, favorite behaviors, consultation behaviors, and negative feedbacks, optimizes the basic feature vectors based on the association behaviors, and outputs optimized feature vectors.

[0087] S123: The recognition layer generates implicit behavior vectors and abnormal behavior vectors based on browsing behaviors, favorite behaviors, consultation behaviors, and negative feedbacks.

[0088] S124: The output layer generates historical behavior features based on the optimized feature vectors, implicit behavior vectors, and abnormal behavior vectors.

[0089] In this embodiment, the quantization layer is the first-layer module in the feature extraction model, which is used to perform basic feature extraction on browsing behaviors and favorite behaviors; the basic feature vectors are the preliminary feature representations output by the quantization layer, including the quantization statistical results of the user's basic behaviors (such as browsing and favoriting); the association layer is the second-layer module in the feature extraction model, which is used to analyze the potential association relationships between different behavior types; the browsing behavior refers to the operation records of the user viewing design cases or designer homepages on the platform, including dimensions such as browsing objects, times, and durations; the favorite behavior refers to the operation records of the user actively saving specific cases or designers, reflecting the explicit preference selections of the user; the consultation behavior refers to the direct interaction records between the user and the designer, including interactive behaviors such as online consultation and appointment communication; the negative feedback refers to the behavior records of the user expressing dissatisfaction, including negative operations such as unfavoriting and complaint evaluation; the optimized feature vectors are the enhanced feature representations obtained after being processed by the association layer, which not only include basic behavior statistics but also incorporate the association pattern information between behaviors; the recognition layer is the third-layer module in the feature extraction model, which is used to identify deep patterns and abnormal situations from the user's behaviors; the implicit behavior vectors are the feature representations generated by the recognition layer that reflect the user's potential preferences, capturing the secondary preferences and potential needs that are not explicitly expressed but persist in the user's behaviors; the abnormal behavior vectors are the feature representations generated by the recognition layer that reflect the user's abnormal behaviors, which are used to identify sudden or contradictory behavior characteristics that do not conform to the user's regular behavior patterns.

[0090] Exemplarily, the historical behavior information of a certain user on the platform includes:

[0091] Browsing behavior: Viewed 8 Nordic-style cases, 3 industrial-style cases, and the viewing time is greater than the preset time; Favorite behavior: Saved the homepages of 2 Nordic-style designers; Consultation behavior: Contacted 1 designer (Nordic style); Negative feedback: Filed 1 complaint about "overpriced quotes" for a consultation.

[0092] Specifically, the browsing behavior and collection behavior are subjected to feature quantization by a quantization layer to generate a basic feature vector. The correlation layer analyzes the correlation behavior among the browsing behavior, collection behavior, consultation behavior, and negative feedback to output an optimized feature vector. Combining the implicit behavior vector and abnormal behavior vector generated by the recognition layer, the historical behavior features are finally fused and output by the output layer, realizing multi-dimensional modeling of user behavior data.

[0093] In one embodiment, as Figure 3 shown, step S123 includes the steps:

[0094] S1231A: The recognition layer performs text analysis on the browsing behavior and collection behavior, and extracts the preference style label and urgency label;

[0095] S1232A: Calculate the preference intensity of the extracted preference style label and the urgency intensity of the urgency label;

[0096] S1233A: Generate an implicit behavior vector based on the preference intensity and the urgency intensity.

[0097] In this embodiment, the text analysis is a process of performing natural language processing on the text information in the user's browsing and collection content (such as case titles, designer tags, description copywriting, etc.) to extract key semantic features; the preference style label is the design style category identified through text analysis (such as "modern minimalist", "neo-Chinese", "industrial style", etc.), which is used to quantify the user's aesthetic preference direction; the urgency label is the time-sensitive feature extracted from the behavior text (such as "urgent", "as soon as possible", "before the end of the month", etc.), which reflects the timeliness requirement of the user's needs; the preference intensity is the quantization value of the preference degree for a specific style label, which is calculated based on data such as the frequency and duration of the style-related behavior; the urgency intensity is the quantization evaluation value for the time-sensitive demand, which is generated by combining features such as the occurrence frequency and text emphasis degree of the urgency label; the implicit behavior vector is a two-dimensional feature representation composed of the preference intensity and the urgency intensity, which reflects the potential demand characteristics not explicitly expressed by the user;

[0098] Exemplarily, a user's browsing behavior on the platform: "Modern minimalist three-bedroom apartment" (stayed for 8 minutes) and "Urgent! Minimalist apartment to be completed in two weeks" (stayed for 11 minutes); collection behavior: collecting the homepage of Designer Wang (whose homepage includes tags: "modern style", "efficient construction") and collecting the homepage of Designer Li (whose homepage includes the case: "48-hour quick installation plan");

[0099] Text analysis process: Extract style tags: Modern Minimalism (appeared 2 times) and Minimalist (appeared 1 time), extract urgency tags: "Urgent! Complete within two weeks" and "48-hour quick installation"; Intensity calculation process: Preference intensity: Modern Minimalism: 0.8 (based on 2 views + 1 favorite); Minimalist: 0.6; Urgency intensity: 0.9 (high-frequency occurrence of time limit words); Implicit behavior vector generation: {Style preference: {"Modern Minimalism": 0.8, "Minimalist": 0.6}, Time urgency: 0.9}, this implicit behavior vector can indicate that the user has a strong preference for Modern Minimalism and has high requirements for construction efficiency.

[0100] Specifically, through the recognition layer, text analysis is performed on browsing behavior and favorite behavior to extract preference style tags and urgency tags, and preference intensity and urgency intensity are calculated to generate an implicit behavior vector, realizing the excavation of users' potential design needs and decision-making psychology.

[0101] In one embodiment, as Figure 4 shown, step S123 includes the steps:

[0102] S1231B: Extract the contradictory behavior characteristics in browsing behavior, favorite behavior, consultation behavior, and negative feedback based on preset contradiction detection rules;

[0103] S1232B: Extract the abnormal frequency characteristics in browsing behavior, favorite behavior, consultation behavior, and negative feedback based on preset frequency detection rules;

[0104] S1233B: Generate an abnormal behavior vector based on the contradictory behavior characteristics and abnormal frequency characteristics.

[0105] In this embodiment, the contradiction detection rule is a predefined behavior logic conflict determination criterion for identifying self-contradictory patterns in user behavior. The contradiction detection rules include: style preference conflict (such as frequently browsing modern style but favoriting classical style), inconsistency between words and deeds (such as canceling immediately after consultation), evaluation contradiction (such as giving a good review but canceling the favorite), etc.; The contradictory behavior characteristics are quantitative indicators extracted through the contradiction detection rules, reflecting the degree of logical inconsistency in user behavior, including: style conflict intensity, behavior reversal frequency, evaluation contradiction index, etc.; The frequency detection rule is a preset behavior frequency abnormality determination criterion for identifying behavior mutations that deviate from the user's normal pattern. The behavior content detected by the frequency detection rule includes: sudden high-frequency behavior, abnormal behavior interval, abnormal time period distribution, etc.; The abnormal frequency characteristics are quantitative indicators extracted through the frequency detection rule, reflecting the degree of unconventionality of behavior frequency, including: frequency deviation value, behavior density index, time period abnormality degree, etc.; The abnormal behavior vector is a comprehensive abnormal indicator composed of contradictory behavior characteristics and abnormal frequency characteristics, used to characterize the abnormal and atypical pattern of user behavior;

[0106] Exemplarily, a user's browsing behavior on the platform: 15 modern minimalist cases, and the browsing time for each is greater than the preset time; favorite behavior: favoriting the homepages of 3 industrial-style designers; consultation behavior: consulting 2 industrial-style designers; negative feedback: canceling the consultation with 1 industrial-style designer midway; their historical behavior pattern (comparison benchmark): average in the past three months: browsing 10 modern minimalist cases per week, favoriting 1-2 modern minimalist designers, and rarely contacting industrial style; contradiction detection process: mainly browsing modern minimalist (15 times) but concentrating on favoriting / consulting industrial style (3 times); contradiction characteristics: style conflict value: 0.8 (full score 1), behavior reversal rate: 0.6; frequency detection process finds that: the contact times of industrial style increase suddenly by 300%, and consultation is made during non-active hours (early morning); abnormal characteristics: frequency deviation degree: 0.9, time period abnormal value: 0.7; abnormal behavior vector generation: {contradiction index: 0.8, frequency abnormality degree: 0.9, comprehensive abnormal value: 0.85}.

[0107] Specifically, the contradiction detection rules set in advance are used to extract the contradiction behavior characteristics in the browsing behavior, favorite behavior, consultation behavior, and negative feedback. At the same time, the frequency detection rules are used to extract the abnormal frequency characteristics in the browsing behavior, favorite behavior, consultation behavior, and negative feedback. Finally, the abnormal behavior vector is generated by fusion to realize the recognition of potential abnormal patterns in the user behavior data.

[0108] In one embodiment, the matching model includes an embedding layer, an interaction layer, and an output layer. As Figure 5 shown, step S30 includes the steps of:

[0109] S31: The embedding layer maps the requirement features, historical behavior features, and multi-dimensional feature vectors to the same embedding space to generate a user embedding vector and a designer embedding vector.

[0110] S32: The interaction layer identifies the interaction correlation between the user embedding vector and the designer embedding vector to generate an interaction feature vector.

[0111] S33: The output layer generates a matching degree matrix based on the interaction feature vector.

[0112] In this embodiment, the embedding layer is the first-layer module of the matching model, which is used to map heterogeneous features to a unified low-dimensional vector space. The embedding layer realizes dimension unification and semantic alignment through non-linear transformation. The user embedding vector is the user feature representation output by the embedding layer, which is generated by fusing demand features and historical behavior features, including explicit demands (such as budget, style preference) and implicit features (such as behavior patterns, potential preferences). The designer embedding vector is the designer feature representation output by the embedding layer, which is obtained by converting multi-dimensional feature vectors, including professional dimensions (such as style expertise, case quality) and service dimensions (such as response speed, quotation range), etc. The interaction layer is the second-layer module of the matching model, which mines through a deep network: vector dot product (calculating the basic matching degree), cross network (capturing non-linear interaction relationships), and attention mechanism (identifying key matching dimensions). The interaction feature vector is the high-order feature representation output by the interaction layer, including explicit interaction features (such as style matching degree) and implicit correlation features (such as potential association between behavior and case). The output layer is the third-layer module of the matching model, which calculates the final matching score through a fully connected network and generates an N×M-dimensional matching degree matrix (N users × M designers). The matching degree matrix is the structured result output by the matching model.

[0113] Specifically, the demand features, historical behavior features, and designer multi-dimensional feature vectors are uniformly mapped to the same embedding space through the embedding layer to generate user embedding vectors and designer embedding vectors. Then, the interaction layer analyzes the interaction relationship between the two to generate interaction feature vectors. Finally, the output layer outputs the matching degree matrix to achieve the deep semantic matching between user demands and designer features.

[0114] In one embodiment, as Figure 6 shown, step S40 includes the steps:

[0115] S41: Based on the matching degree matrix, match the preset corresponding matching error loss function and diversity constraint loss function;

[0116] S42: Based on the matched matching error loss function and diversity constraint loss function, construct a multi-objective loss function and match the associated solution algorithm;

[0117] S43: Generate a recommended solution set based on the multi-objective loss function;

[0118] S44: Generate several demand solutions for the recommended solution set based on the matched solution algorithm.

[0119] In this embodiment, the matching error loss function is a function used to evaluate the difference between the predicted matching degree and the ideal matching degree. Commonly used ones include mean squared error (measuring the accuracy of score prediction), cross-entropy loss (dealing with binary classification matching problems), and contrastive loss (enhancing the distinguishability between positive and negative samples); the diversity constraint loss function is a penalty term to ensure the diversity of the recommendation results. Common types include category distribution entropy (promoting uniform coverage in dimensions such as style / price), similarity penalty (suppressing the similarity of designers in the recommendation list), and long-tail compensation (enhancing the exposure opportunities of unpopular designers); the multi-objective loss function is a composite objective function composed of multiple sub-losses weighted; the solution algorithm is a mathematical method used to minimize the loss function, including gradient descent algorithms, evolutionary algorithms, and heuristic algorithms (simulated annealing); the set of recommended solutions is the set of candidate recommendation schemes generated during the solution process, including: recommendation lists under different weight configurations, schemes with different diversity levels, etc.; the solution to be obtained is the recommended scheme after final screening.

[0120] Specifically, the joint optimization of the matching error loss function and the diversity constraint loss function is triggered through the matching degree matrix, a multi-objective loss function is constructed, and a set of recommended solutions is generated. Finally, several solutions to be obtained are output through the solution algorithm, achieving an intelligent balance between the accuracy and diversity of the recommendation results.

[0121] It should be understood that the magnitudes of the sequence numbers of the steps in the above embodiments do not mean the order of execution. The order of execution of each process should be determined according to its function and internal logic, and should not constitute any limitation to the implementation process of the embodiments of this application.

[0122] In one embodiment, an intelligent home improvement designer recommendation system is provided. This intelligent home improvement designer recommendation system corresponds one-to-one with the intelligent home improvement designer recommendation method in the above embodiment. This intelligent home improvement designer recommendation system includes:

[0123] A feature extraction module, configured to obtain user demand information and historical behavior information, and extract demand features and historical behavior features respectively;

[0124] A classification module, configured to obtain a designer database, and classify the designer information in the designer database based on a pre-set classification algorithm to correspondingly form multi-dimensional feature vectors associated with the designer information;

[0125] An input module, configured to input the demand features, historical behavior features, and multi-dimensional feature vectors into the same pre-trained matching model to generate a matching degree matrix;

[0126] A solution generation module for generating several solutions to be obtained based on a predefined loss function and the solution algorithm associated with the loss function when receiving the matching degree matrix output by the matching model;

[0127] A recommended list generation module, configured to generate a recommended list based on a plurality of generated requirement solutions and send the recommended list to the client that sends the user requirement information.

[0128] For the specific limitations of a home improvement designer intelligent recommendation system, reference may be made to the limitations of a home improvement designer intelligent recommendation method in the foregoing text, which will not be elaborated herein. Each module in the above-mentioned home improvement designer intelligent recommendation system can be implemented in whole or in part by software, hardware, and their combination. The above-mentioned modules can be embedded in the processor of the computer device in hardware form or independent of it, or stored in the memory of the computer device in software form, so as to facilitate the processor to call and execute the operations corresponding to the above-mentioned modules.

[0129] The above-described embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that: they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements on some of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present application, and should all be included in the protection scope of the present application.

Claims

1. An intelligent recommendation method for home decoration designers, characterized in that: Including the steps: Obtain user requirement information and historical behavior information, and respectively extract requirement features and historical behavior features; Obtain a designer database, and classify the designer information in the designer database based on a preset classification algorithm to correspondingly form multi-dimensional feature vectors associated with the designer information; Input the requirement features, historical behavior features, and multi-dimensional feature vectors into the same pre-trained matching model to generate a matching degree matrix; When receiving the matching degree matrix output by the matching model, generate a number of requirement solutions based on a predefined loss function and a solution algorithm associated with the loss function; Generate a recommendation list based on the generated number of requirement solutions and send it to the user terminal that sent the user requirement information.

2. The intelligent recommendation method for home decoration designers according to claim 1, characterized in that: The step of obtaining user requirement information and historical behavior information, and respectively extracting requirement features and historical behavior features includes the steps: Identify the identity information corresponding to the user requirement information, and obtain the historical behavior information of the identity information; Input the historical behavior information into a pre-trained feature extraction model, so that the feature extraction model extracts and outputs historical behavior features based on the historical behavior information.

3. The intelligent recommendation method for a home decoration designer according to claim 2, characterized in that: The feature extraction model includes a quantization layer, an association layer, an identification layer, and an output layer. The historical behavior information includes browsing behavior, collection behavior, consultation behavior, and negative feedback. The step of inputting the historical behavior information into a pre-trained feature extraction model, so that the feature extraction model extracts and outputs historical behavior features based on the historical behavior information includes the steps: The quantization layer quantizes the features of the browsing behavior and the collection behavior and outputs a basic feature vector; The association layer captures the association behaviors among the browsing behavior, the collection behavior, the consultation behavior, and the negative feedback, and optimizes the basic feature vector based on the association behaviors to output an optimized feature vector; The identification layer generates an implicit behavior vector and an abnormal behavior vector based on the browsing behavior, the collection behavior, the consultation behavior, and the negative feedback; The output layer generates historical behavior features based on the optimized feature vector, the implicit behavior vector, and the abnormal behavior vector.

4. The intelligent recommendation method for a home decoration designer according to claim 3, wherein: The step of the identification layer generating an implicit behavior vector and an abnormal behavior vector based on the browsing behavior, the collection behavior, the consultation behavior, and the negative feedback includes the steps: The identification layer performs text analysis on the browsing behavior and the collection behavior, and extracts a preference style label and an urgency label; Calculate the preference intensity of the extracted preference style label and the urgency intensity of the urgency label; Generate an implicit behavior vector based on the preference intensity and the urgency intensity.

5. The intelligent recommendation method for a home decoration designer according to claim 3, wherein: The step of the identification layer generating an implicit behavior vector and an abnormal behavior vector based on the browsing behavior, the collection behavior, the consultation behavior, and the negative feedback includes the steps: Extract the contradictory behavior features in the browsing behavior, the collection behavior, the consultation behavior, and the negative feedback based on a preset contradiction detection rule; Extract the abnormal frequency features in the browsing behavior, the collection behavior, the consultation behavior, and the negative feedback based on a preset frequency detection rule; Generate an abnormal behavior vector based on the contradictory behavior features and the abnormal frequency features.

6. The intelligent recommendation method for a home decoration designer according to claim 1, wherein: The matching model includes an embedding layer, an interaction layer, and an output layer. The step of inputting the requirement feature, historical behavior feature, and multi-dimensional feature vector into the same pre-trained matching model to generate a matching degree matrix includes the steps: The embedding layer maps the requirement feature, historical behavior feature, and multi-dimensional feature vector to the same embedding space to generate a user embedding vector and a designer embedding vector; The interaction layer identifies the interaction correlation between the user embedding vector and the designer embedding vector to generate an interaction feature vector; The output layer generates a matching degree matrix based on the interaction feature vector.

7. The intelligent recommendation method for a home decoration designer according to claim 1, wherein: The step of generating a number of demand solutions based on a predefined loss function and a solution algorithm associated with the loss function when receiving the matching degree matrix output by the matching model includes the steps: Match a preset corresponding matching error loss function and diversity constraint loss function based on the matching degree matrix; Construct a multi-objective loss function based on the matched matching error loss function and diversity constraint loss function, and match the associated solution algorithm; Generate a recommended solution set based on the multi-objective loss function; Generate a number of demand solutions for the recommended solution set based on the matched solution algorithm.

8. An intelligent recommendation system for home decoration designers, characterized in that: Including: A feature extraction module for obtaining user requirement information and historical behavior information, and respectively extracting requirement features and historical behavior features; A classification module for obtaining a designer database and classifying the designer information in the designer database based on a preset classification algorithm to correspondingly form a multi-dimensional feature vector associated with the designer information; An input module for inputting the requirement feature, historical behavior feature, and multi-dimensional feature vector into the same pre-trained matching model to generate a matching degree matrix; A demand solution generation module for generating a number of demand solutions based on a predefined loss function and a solution algorithm associated with the loss function when receiving the matching degree matrix output by the matching model; A recommended list generation module for generating a recommended list based on the generated number of demand solutions and sending it to the user terminal that sent the user requirement information.

Images