Method, device and equipment for material search

Abstract

The application provides a material searching method, device and equipment. In the method, long-term preference characteristics of a user are obtained when a scheme is designed, material characteristics of each candidate material are obtained after the candidate material meeting the search condition is searched, instant search characteristics of the user are generated according to the search condition, instant preference characteristics of the user when designing the target scheme are determined according to data of the target scheme, and the long-term preference characteristics, the instant preference characteristics, the instant search characteristics and the material characteristics of each candidate material are used to accurately predict probability information of the user selecting each candidate material through a selection probability estimation model, and the candidate material with higher selection probability is recommended to the user according to the probability information of the user selecting each candidate material, so that the search result is closer to the search intention of the user, the accuracy of the material searching is improved, and the time consumed by the designer user in searching for the required material is shortened and the efficiency is improved.

Description

Technical Field

[0001] This application relates to the field of computer technology, and in particular to a method, apparatus and device for material searching. Background Technology

[0002] In interior design and other scenarios, when using 3D design tools to design schemes, designers need to search for materials that meet the client's apartment layout and decoration style requirements from the millions of materials (i.e., object models, such as sofas, lamps, etc.) provided by the 3D design tools, and then match and arrange them. The designer's search for materials is characterized by long time sequence, high frequency, and heavy matching.

[0003] Current 3D design tools typically allow users to explicitly set search criteria to filter materials. However, the results of explicit filtering usually still contain a huge number of materials, requiring designers to manually search for the required materials from a large pool of data. This results in low accuracy of material search results and is time-consuming and inefficient for designers. Summary of the Invention

[0004] This application provides a method, apparatus, and device for material searching, which solves the problems of low accuracy of material search results in 3D design scenarios and the long time and low efficiency for designers to find the required materials.

[0005] Firstly, this application provides a method for material searching, including:

[0006] Receive material search requests triggered by users during the design of target solutions, and obtain search conditions, long-term preference characteristics of the user in solution design, and data of the target solution;

[0007] Search for candidate materials that meet the search criteria, and obtain the material characteristics of each candidate material; generate the user's real-time search characteristics based on the search criteria; and determine the user's real-time preference characteristics when designing the target solution based on the data of the target solution.

[0008] Based on the user's long-term preference characteristics, immediate preference characteristics, immediate search characteristics, and the material characteristics of each candidate material, predict the probability information of the user selecting each candidate material;

[0009] Based on the probability information of the user selecting each of the candidate materials, the search results are output.

[0010] Secondly, this application provides a method for material searching, including:

[0011] Receive material search requests triggered by users during the design of target solutions, obtain search conditions, and obtain the user's design behavior representation information and the user's material selection timing behavior representation information when designing target solutions;

[0012] After finding candidate materials that meet the search criteria, material recommendations are made based on the user's design behavior representation information and the user's material selection time sequence behavior representation information when designing the target scheme, so as to recommend candidate materials that match the user's design behavior preferences.

[0013] Thirdly, this application provides a material searching apparatus, comprising:

[0014] The data acquisition module is used to receive material search requests triggered by users during the design of target solutions, and to acquire search conditions, long-term preference characteristics of the user in solution design, and data of the target solution.

[0015] The search module is used to search for candidate materials that meet the search criteria;

[0016] The feature acquisition module is used to acquire the material features of each candidate material; generate the user's real-time search features based on the search conditions; and determine the user's real-time preference features when designing the target solution based on the data of the target solution.

[0017] A probability prediction module is selected to predict the probability information of the user selecting each of the candidate materials based on the user's long-term preference characteristics, immediate preference characteristics, immediate search characteristics, and material characteristics of each candidate material.

[0018] The search result processing module is used to output search results based on the probability information of the user selecting each of the candidate materials.

[0019] Fourthly, this application provides a material searching apparatus, comprising:

[0020] The behavior representation module is used to receive material search requests triggered by users during the design of target solutions, obtain search conditions, as well as the user's design behavior representation information and the material selection timing behavior representation information when the user designs target solutions.

[0021] The material search module is used to recommend candidate materials that match the user's design behavior preferences to the user after finding candidate materials that meet the search conditions, based on the user's design behavior representation information and the user's material selection time sequence behavior representation information when designing the target scheme.

[0022] Fifthly, this application provides an electronic device, including: a processor, and a memory communicatively connected to the processor;

[0023] The memory stores computer-executed instructions;

[0024] The processor executes computer execution instructions stored in the memory to implement the method described in the first or second aspect above.

[0025] In a sixth aspect, this application provides a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, are used to implement the method described in the first or second aspect above.

[0026] The material search method, apparatus, and equipment provided in this application, when searching for materials based on user requests, acquire the user's long-term preference characteristics for scheme design, perform explicit user filtering based on search conditions, and after finding candidate materials that meet the search conditions, acquire the material characteristics of each candidate material, generate the user's real-time search characteristics based on the search conditions, determine the user's real-time preference characteristics when designing the target scheme based on the target scheme data, and predict the probability information of the user selecting each candidate material through the selection probability prediction model based on the user's long-term preference characteristics, real-time preference characteristics, real-time search characteristics, and the material characteristics of each candidate material, thus accurately predicting the probability information of the user selecting each candidate material. Furthermore, based on the probability information of the user selecting each candidate material, candidate materials with higher selection probabilities are recommended to the user, making the search results closer to the user's search intent, improving the accuracy of material search, thereby reducing the time spent by designers in finding the required materials and improving efficiency. Attached Figure Description

[0027] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0028] Figure 1 Example diagram of the system architecture on which the material search method provided in this application is based;

[0029] Figure 2 A flowchart illustrating a material search method provided in an exemplary embodiment of this application;

[0030] Figure 3 A flowchart of a material search method provided for another exemplary embodiment of this application;

[0031] Figure 4 A flowchart of a material search method provided for applying for another exemplary embodiment;

[0032] Figure 5 A schematic diagram of the structure of a material searching apparatus provided for an exemplary embodiment of this application;

[0033] Figure 6 A schematic diagram of the structure of a material searching apparatus provided for an exemplary embodiment of this application;

[0034] Figure 7 This is a schematic diagram of the structure of an electronic device provided in an example embodiment of this application.

[0035] The accompanying drawings illustrate specific embodiments of this application, which will be described in more detail below. These drawings and descriptions are not intended to limit the scope of the concept in any way, but rather to illustrate the concept of this application to those skilled in the art through reference to particular embodiments. Detailed Implementation

[0036] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application as detailed in the appended claims.

[0037] First, let me explain the terms used in this application:

[0038] Factorization Machines (FM) are one of the classic models for predicting Click-Through Rate (CTR).

[0039] Explicit user filtering: A search method where users combine and filter material attributes and sorting criteria provided on the interactive page.

[0040] Implicit system filtering: The process by which the system filters and sorts materials based on user-defined search criteria.

[0041] To address the issues of current 3D design tools, where explicit user search results often contain a vast number of materials, requiring designers to manually search for the desired materials, resulting in low accuracy and time-consuming, inefficient searches, this application provides a material search method. Upon receiving a material search request triggered by a user during the design process, the method acquires search criteria, the user's long-term design preferences, and data on the target solution. Based on the search criteria, it searches for candidate materials that meet those criteria, thus achieving explicit user filtering. Based on the candidate materials selected by the user, the system acquires the material characteristics of each candidate material; generates the user's real-time search characteristics according to the search criteria; and determines the user's real-time preference characteristics when designing the target solution based on the target solution data. The user's long-term preference characteristics, real-time preference characteristics, real-time search characteristics, and the material characteristics of each candidate material are input into a trained selection probability prediction model. This model predicts the probability of the user selecting each candidate material. Based on the probability of the user selecting each candidate material, the system filters out the candidate materials with higher selection probabilities as the final search results. This makes the search results more consistent with the user's long-term and short-term material selection intentions, improving the accuracy of the material search results. Consequently, it reduces the time spent by designers in finding the required materials and increases efficiency.

[0042] The material search method provided in this application can be used in model search scenarios in 3D design tools, or in other similar material search scenarios.

[0043] For example, the material search method provided in this application can be applied to Figure 1 The system architecture is shown below. Figure 1 As shown, the system architecture includes: terminals and servers.

[0044] The server can be the server hosting the 3D design tool, specifically a server cluster deployed in the cloud, such as a 3D cloud design tool. This server hosts the 3D design tool and stores user-related assets (including design schemes, materials, coupons, etc.) and behavior logs (such as logs of user actions like uploading materials, collecting, following, copying, sharing schemes or materials, and rendering design schemes). The server also stores a material library and a selection probability prediction model. Through pre-set computational logic, the server predicts the probability of a user selecting candidate materials based on the user's long-term preference characteristics, immediate preference characteristics, immediate search characteristics, and the material characteristics of the candidate materials, and determines the candidate materials with the higher probability of selection as the search result.

[0045] The terminal can specifically be a hardware device with network communication, computing and information display functions, including but not limited to smartphones, tablets, desktop computers, and Internet of Things devices.

[0046] Through communication with the server, when a user searches for materials while designing a target solution using the terminal, the terminal sends a material search request to the server. Upon receiving the request, the server obtains the search criteria, user information, and target solution information. Based on the user information, it retrieves the user's long-term preference characteristics for solution design. It then searches for candidate materials that meet the search criteria; optionally, it can also search for candidate materials that match the user's long-term preference characteristics. The server obtains the material characteristics of each candidate material and generates the user's immediate search characteristics based on the search criteria. Based on the target solution data, it determines the user's immediate preference characteristics when designing the target solution. The server inputs the user's long-term preference characteristics, immediate preference characteristics, immediate search characteristics, and the material characteristics of each candidate material into a trained selection probability prediction model. The model predicts the probability of the user selecting each candidate material. Based on the probability of the user selecting each candidate material, the server determines the search results. The server then returns the search results to the terminal. The terminal displays the search results so that the user can select the materials they need for solution design.

[0047] The technical solution of this application and how the technical solution of this application solves the above-mentioned technical problems are described in detail below with specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments. The embodiments of this application will now be described with reference to the accompanying drawings.

[0048] Figure 2 This is a flowchart illustrating a material search method provided for an exemplary embodiment of this application. The execution entity in this embodiment may be the aforementioned server, such as... Figure 2 As shown, the specific steps of this method are as follows:

[0049] Step S201: Receive the material search request triggered by the user during the design of the target solution, and obtain the search conditions, the user's long-term preference characteristics for solution design, and the target solution data.

[0050] The material search request includes search criteria, user information, and information about the target solution that the user is currently designing.

[0051] Upon receiving a physical search request, the request can be parsed to obtain the search criteria, user information, and target solution information.

[0052] Search criteria are the filter conditions set by users when searching for materials through an interactive page. They are the filter conditions used by users to select materials in the material library.

[0053] User information can include unique identifiers such as user ID, user account, and mobile phone number. Based on this information, the user's long-term preference characteristics can be retrieved from offline-generated long-term user preference features.

[0054] In this embodiment, the long-term preference characteristics of users in designing solutions (hereinafter referred to as the long-term preference characteristics of users) can be generated and stored offline by the server. The server generates and stores the long-term preference characteristics of each user offline based on the design-related assets and behavioral data of each user over a long historical period.

[0055] For example, the long-term preference features of each user can be stored in a key-value pair manner, where the key can be the user's identification information and the value is the user's long-term preference feature.

[0056] The target solution's information can be its identifying information, such as its solution ID or solution name, which uniquely identifies the target solution. Based on this information, the target solution's data can be retrieved.

[0057] In addition, in the cold start scenario, for new users, the long-term preference characteristics of popular users can be transferred to new users as long-term preference characteristics of new users, thus solving the cold start problem.

[0058] Step S202: Search for candidate materials that meet the search criteria and obtain the material characteristics of each candidate material.

[0059] After obtaining the search criteria, a material search can be performed based on the search criteria to recall materials that meet the search criteria from the material library as candidate materials.

[0060] After finding candidate materials, the material characteristics of each candidate material are obtained. These material characteristics may include the material's label information, its category, and its attribute information.

[0061] In 3D design scenarios, the materials used (i.e., item models) are typically categorized into multiple classes, and the category to which each item belongs can be set in the material library. Materials usually have various attribute information such as style, brand, color, and origin. In addition, the material library also sets tags for each material, including system tags and custom tags. For example, tag information such as material, shape, texture, and color.

[0062] Step S203: Generate the user's real-time search characteristics based on the search criteria.

[0063] The search criteria include constraint objects and constraint conditions, and the constraint conditions include condition values.

[0064] One optional implementation of this step is to convert the tag-type condition values ​​(including material category, attributes, and tag information) in the search criteria into numerical features based on the set index dictionary. The user's immediate search features include the numerical features converted from the tag-type condition values ​​in the search criteria, as well as the numerical condition values ​​in the search criteria.

[0065] Another optional implementation of this step is to use word embedding technology to vectorize the search conditions to obtain the feature vectors corresponding to the search conditions, which can then be used as the user's immediate search features.

[0066] In addition, this step can be combined with the two implementation methods mentioned above and / or other implementation methods to obtain more features, which will not be elaborated here.

[0067] Step S204: Based on the data of the target solution, determine the user's immediate preference characteristics when designing the target solution.

[0068] In this embodiment, based on the data of the target scheme designed by the user, the scheme feature information such as the scheme area, number of spaces, creation method, and creation environment of the target scheme is extracted. Based on the selection sequence of the materials selected in the target scheme, the user's attribute preference features for each category are statistically analyzed. The scheme feature information and the user's attribute preference features for each category are used as the user's real-time preference features when designing the target scheme, which can represent the user's current search intent and preferences.

[0069] In this embodiment, steps S202-S204 can be performed in parallel, and there is no specific limitation on the execution order of these three steps.

[0070] In this embodiment, long-term user preference features are generated based on the user's design behavior over a longer historical period. Considering that user interest in materials changes with search intent and time, real-time user preference features are generated based on the data of the user's current design target solution. Combining long-term and real-time preference features can better characterize user preferences across different search intents and time ranges. Furthermore, combining real-time search features determined based on search conditions ensures that the set of materials exposed to users in real-world application scenarios more closely matches the user's current search intent. By comprehensively utilizing data from multiple dimensions, such as design-related asset data, behavior logs, and tagging systems, to digitally represent the designer's design behavior, more multi-dimensional feature information is obtained, which can better represent the designer's true intent in material searches.

[0071] For example, after a user (designer) completes a design for a large apartment in a light luxury style, they then begin designing a design for a small apartment in a country style. The user's long-term preferences remain consistent throughout the design process of both designs. However, their immediate preference characteristics are recalculated based on the model selection sequence in the current target design and the information (Meta information) of the design itself. Therefore, when the user's design intent changes, the material search method provided in this application can quickly characterize the design behavior after the change, forming different immediate preference characteristics to better represent the user's current search intent and perform reasonable material search and recommendation.

[0072] Step S205: Based on the user's long-term preference characteristics, immediate preference characteristics, immediate search characteristics, and the material characteristics of each candidate material, predict the probability information of the user selecting each candidate material.

[0073] After obtaining the user's long-term preference features, immediate preference features, immediate search features, and material features of each candidate material, the material features of each candidate material are input together with the user's long-term preference features, immediate preference features, and immediate search features into the trained selection probability prediction model. The selection probability prediction model then predicts the probability information of the user selecting each candidate material, that is, the selection probability information of each candidate material.

[0074] Among them, the probability information of users selecting candidate materials is the probability information of the candidate material being selected after it is output as a search result. It is used to measure the likelihood of the candidate material being selected by the user.

[0075] For example, a user's selection behavior can be clicking on a material, and the selection probability prediction model can be a click-through rate (CTR) prediction model, with the predicted probability information being the probability information of the user clicking on the candidate material.

[0076] The selected probability prediction model can be obtained by training a deep learning model.

[0077] For example, the probabilistic prediction model can be a deep learning model that combines shallow and deep models, such as DeepFM, Probabilistic Neural Network (PNN), NFM, and Wide&Deep. By combining shallow models (such as the FM layer of DeepFM) and deep models (such as the Deep layer of DeepFM), the relationship between higher-order behavioral representations and material selection can be explored, so that new materials can also get a fairer exposure opportunity in the lead-up scenario.

[0078] Step S206: Output the search results based on the probability information of the user selecting each candidate material.

[0079] After determining the probability information of the user selecting each candidate material, the candidate material with the higher probability of selection is selected as the search result and the search result is output.

[0080] Optionally, based on the probability information of the user selecting each candidate material, information on candidate materials whose probability information is greater than or equal to a probability threshold is used as search results and recommended to the user. The probability threshold can be set and adjusted according to the actual application scenario, and is not specifically limited here.

[0081] Optionally, the candidate materials are sorted based on the probability information of the user selecting each candidate material. A specified number of candidate materials are then presented to the user as search results based on the sorting results. The specified number can be set and adjusted according to the actual application scenario; no specific limitation is made here.

[0082] In addition, when outputting search results, the search results can be displayed in pages according to the offset set by the user, where the offset is the number of materials displayed on each page.

[0083] This embodiment models the material search problem as a material click probability prediction and ranking problem based on the design behavior representations of different users (users' long-term preference characteristics, immediate preference characteristics, and immediate search characteristics). This enables the materials exposed to users to better match their true material search intentions, thereby improving user satisfaction with material search results and greatly helping to improve design efficiency.

[0084] In this embodiment, when searching for materials based on user requests, the system acquires the user's long-term preference characteristics for designing solutions. After explicit user filtering based on search criteria and finding candidate materials that meet the criteria, it acquires the material characteristics of each candidate material. Based on the search criteria, it generates the user's immediate search characteristics and determines the user's immediate preference characteristics when designing the target solution based on the target solution data. Then, based on the user's long-term preference characteristics, immediate preference characteristics, immediate search characteristics, and the material characteristics of each candidate material, it uses a selection probability prediction model to predict the probability of the user selecting each candidate material, accurately predicting the probability of selection. Furthermore, based on the probability of selection, the system recommends candidate materials with higher selection probabilities to the user, making the search results closer to the user's search intent, improving the accuracy of material searches, and thus reducing the time and increasing the efficiency for designers to find the materials they need.

[0085] In one optional embodiment, long-term preference features of users for scheme design (hereinafter referred to as long-term preference features of users) can be generated offline based on data such as design-related asset data and material search log data of users over a long historical period, so as to improve the efficiency of online material search.

[0086] For example, generating long-term user preference features can be achieved through the following steps S1-S2:

[0087] Step S1: Obtain the user's design-related asset data and material search log data within a preset historical time period.

[0088] The preset historical period is a relatively long time frame that includes a significant amount of user design behavior data. For example, the preset historical period could be the past 3 months, 180 days, or one year, and can be set and adjusted according to the actual application scenario. No specific limitations are set here.

[0089] Design-related asset data includes asset data related to user design behavior, such as data on design schemes created by users, data on user behaviors such as rendering, sharing, following, copying, and collecting schemes, data on user purchases of coupons for rendering, data on materials on users, and data on user behaviors such as sharing, following, copying, and collecting materials.

[0090] The material search log data records the user's search behavior for materials, including information on materials that have been exposed to (recommended) to the user based on the user's search behavior, as well as information on whether the user has selected them.

[0091] Step S2: Determine the long-term preference characteristics of each user based on the design-related asset data and material search log data of each user within a preset historical period.

[0092] In this embodiment, long-term preference characteristics include asset distribution characteristics and category attribute preference characteristics.

[0093] The asset distribution characteristics include at least one of the following: total number of schemes, average number of spaces per scheme, average number of saves per scheme, average number of renders per scheme, percentage of selected commodity materials, and percentage of selected public warehouse materials.

[0094] Category attribute preference features include at least one of the following: the number of materials selected by the user for each attribute under each category, and the user's preferred material attributes.

[0095] The total number of schemes refers to the total number of schemes designed by the user within the preset historical period. The average number of spaces per scheme refers to the average number of spaces contained in all schemes designed by the user within the preset historical period. The average number of renderings per scheme refers to the average number of renderings of all schemes designed by the user within the preset historical period.

[0096] In a 3D design scenario, materials refer to item models, which are divided into product models purchased by users and public library models provided by the public library (a public material library).

[0097] The proportion of selected commodity materials refers to the percentage of commodity materials among all materials selected by the user in all schemes designed within a preset historical period. The proportion of selected public warehouse materials refers to the percentage of public warehouse materials among all materials selected by the user in all schemes designed within a preset historical period.

[0098] For example, the material attributes preferred by users may include at least one of the following: the material attribute with the most selected materials; the material attribute with the most selected times; the material attribute with the most selected materials (i.e., the material attribute with a large number of selected materials); and the material attribute with the most selected times (i.e., the material attribute with a large number of selected materials).

[0099] For example, in a 3D design scenario, the following primary categories can be set: Structure, Hard Furnishings, Furniture, Kitchen & Bathroom, Lighting & Appliances, Home Decor, and Commercial Decoration. Material attributes can include the following four types: Style, Brand, Color, and Source. In this embodiment, for each material under each primary category, the attributes most frequently used by users (i.e., user preferences) are statistically analyzed for each material attribute. For example, for each category (primary category), the user's preferred style (one or more styles), brand, color, and source are statistically analyzed. User preference attributes may differ across categories; for example, different categories may have different user preferences for styles.

[0100] In addition, each first-level category can also be set with second-level and third-level categories. In other embodiments, feature information can also be calculated for the second-level or third-level categories.

[0101] This embodiment calculates the user's long-term preference characteristics for scheme design based on the user's design-related asset data and material search log data within a preset historical period. This can accurately characterize the user's long-term preferences for materials selected in 3D design, providing more dimensional feature references for material search and improving the accuracy of material search.

[0102] Figure 3This is a flowchart illustrating a material search method as provided in another exemplary embodiment of this application. Based on the above method embodiments, in this embodiment, the server can also generate query conditions based on long-term preference features and search conditions to search for candidate materials that meet the query conditions. The query conditions are used to search for materials that match the user's long-term preferences.

[0103] like Figure 3 As shown, the specific steps of this method are as follows:

[0104] Step S301: Receive the material search request triggered by the user during the design of the target solution, and obtain the search conditions, the user's long-term preference characteristics for solution design, and the target solution data.

[0105] This step is implemented in the same way as step S201 above. For details, please refer to the description of step S201, which will not be repeated here.

[0106] Step S302: Search for candidate materials that meet the search criteria.

[0107] After obtaining the search criteria, a material search can be performed based on the search criteria to recall materials that meet the search criteria from the material library as candidate materials.

[0108] Step S303: Generate query conditions based on long-term preference characteristics and search conditions, and search for candidate materials that meet the query conditions.

[0109] In this embodiment, the server can also search for materials that match the user's long-term preferences as candidate materials based on the user's long-term preference characteristics.

[0110] In this step, the search criteria are rewritten based on the user's long-term preference characteristics to generate a new query condition. This query condition is used to search for materials that match the user's long-term preferences.

[0111] Specifically, the target category of the candidate materials found based on the search criteria is determined. Based on the category attribute preference features in the user's long-term preference characteristics, query conditions are generated. These query conditions are used to search for candidate materials with the material attributes that the user prefers under the target category, in order to recall candidate materials that match the user's long-term preferences. This allows the search results to include materials that are closer to the user's long-term preferences, thereby improving the accuracy of the search results and the click-through rate.

[0112] For example, the category attribute preference features in a user's long-term preference characteristics may include the user's most preferred style (e.g., style A), most preferred brand (e.g., brand B), most preferred color (e.g., color C), and most preferred source (e.g., source D). Assuming the target category of the candidate materials found based on the search criteria is "furniture," then materials with style A, brand B, color C, and source D under the "furniture" category can be generated as candidate materials. Furthermore, if there are multiple user-preferred attributes within the same category, then the materials found based on the query criteria will possess any one of the user-preferred attributes within that category.

[0113] The above steps S302 and S303 can be executed in parallel or sequentially. Here, there is no specific limitation on the execution order of steps S302 and S303.

[0114] Step S304: Obtain the material characteristics of each candidate material.

[0115] Specifically, the label information, attribute information, and category of each candidate material are obtained to obtain the material characteristics of each candidate material, so as to accurately characterize the material characteristics.

[0116] Step S305: Generate the user's real-time search characteristics based on the search criteria and query criteria.

[0117] In this embodiment, when generating a user's immediate search features, the user's immediate search features are generated based on the search conditions and the query conditions generated based on long-term preference features and search conditions, so as to mine more multi-dimensional features to characterize the user's current search intent.

[0118] Both search criteria and query criteria include constraint objects and constraint conditions, and constraint conditions include condition values.

[0119] One optional implementation of this step is to encode the tag-type condition values ​​(including material category, attribute, and tag information) in the search and query conditions into corresponding numerical features based on the set index dictionary. The user's instant search features include the numerical features converted from the tag-type condition values ​​in the search and query conditions, as well as the numerical condition values ​​in the search and query conditions.

[0120] Another optional implementation of this step is to use word embedding technology to perform word embedding processing on the search conditions and query conditions to generate corresponding word embedding vectors, and the user's instant search features include the word embedding vectors.

[0121] In another optional implementation of this step, the two implementations described above can be combined. The user's real-time search features include numerical features converted from the condition values ​​of the tag class in the search conditions and query conditions, as well as word embedding vectors corresponding to the search conditions and query conditions, in order to mine more multi-dimensional features that can characterize the user's current search intent.

[0122] In addition, this step can combine any of the above implementation methods with other implementation methods to obtain more features.

[0123] Step S305 is executed after step S303, and step S305 can be executed in parallel with step S304.

[0124] Step S306: Based on the data of the target solution, determine the user's immediate preference characteristics when designing the target solution.

[0125] In this embodiment, based on the data of the target scheme designed by the user, the scheme feature information such as the scheme area, number of spaces, creation method, and creation environment of the target scheme is extracted. Based on the selection sequence of the materials selected in the target scheme, the user's attribute preference features for each category are statistically analyzed. The scheme feature information and the user's attribute preference features for each category are used as the user's real-time preference features when designing the target scheme, which can represent the user's current search intent and preferences.

[0126] Specifically, based on the target solution data, the solution characteristics are determined. Based on the target solution data, the material selection sequence for the target solution is determined. Based on the material selection sequence, the category attribute preference characteristics of the user when designing the target solution are generated. The user's immediate preference characteristics include the solution characteristics of the target solution and the material sequence characteristics of the user when designing the target solution. The material sequence characteristics include: the material selection sequence and / or the category attribute preference characteristics of the user when designing the target solution.

[0127] The scheme features include at least one of the following: total area, number of spaces, creation method, and creation environment information.

[0128] When designing a target solution, the user's category attribute preference features include at least one of the following: the quantity of materials for each attribute under each category selected in the target solution, the user's preferred material attributes in the target solution, and the selection sequence of material attributes for each category under each attribute.

[0129] Total area refers to the sum of the areas of all spaces in the target design. Creation method refers to how the initial design on which the target design is based was created. The initial design on which the target design is based can be a copied design, a previously drawn floor plan, or a floor plan converted from an imported image, etc.

[0130] In practical applications, 3D design tools can provide different design environments for different types of users, and different design environments have different functions. Creating environment information refers to the information about the design environment in which the target solution resides.

[0131] The material selection sequence of the target plan refers to the information of the materials selected in the target plan, arranged in the order in which the materials were selected.

[0132] Specifically, when determining the material selection sequence of the target solution, the material selection sequence that has been saved in the target solution can be generated and merged with the material selection sequence that has been selected but not yet saved in the current session to obtain the material selection sequence of the target solution.

[0133] For example, based on the material selection sequence of the target solution, for each category and each attribute (such as style), the sequence of the selected materials for that category is determined. Furthermore, the material attributes preferred by users in the target solution can also be statistically analyzed.

[0134] For example, based on the material selection sequence of the target solution, for the "style" attribute under the "furniture" category, the "style" attributes of the materials selected under the "furniture" category in the target solution are arranged into a material style attribute sequence according to the order of material selection. Furthermore, the preferred style attributes of users can also be statistically analyzed.

[0135] By determining the characteristics of the target solution based on the data of the target solution, and determining the material selection sequence of the target solution, the category attribute preference characteristics of the user when designing the target solution are generated based on the material selection sequence of the target solution. This generates the user's real-time preference characteristics when designing the target solution, which can better represent the user's material search intent when designing the target solution. Combining the user's long-term preference characteristics can better characterize the user's preference representation in different search intents and time ranges. Furthermore, by combining the real-time search characteristics determined based on search conditions, the set of materials exposed to the user in the actual application scenario is closer to the user's current search intent.

[0136] Step S306 can be executed in parallel with steps S302-S305 above.

[0137] Step S307: Input the user's long-term preference features, real-time preference features, real-time search features, and material features of each candidate material into the trained selection probability prediction model, and predict the probability information of the user selecting each candidate material through the selection probability prediction model.

[0138] After obtaining the user's long-term preference features, immediate preference features, immediate search features, and material features of each candidate material, the material features of each candidate material are input together with the user's long-term preference features, immediate preference features, and immediate search features into the trained selection probability prediction model. The selection probability prediction model then predicts the probability information of the user selecting each candidate material, that is, the selection probability information of each candidate material.

[0139] For example, for each candidate material, the user's long-term preference features, immediate preference features, and immediate search features are concatenated with the material features of the candidate material to obtain concatenated features, thereby achieving the fusion of multi-dimensional features. The concatenated features are then subjected to dimensionality reduction processing, and the dimensionality-reduced features are input into a trained selection probability prediction model. This model predicts the probability information of the user selecting the candidate material. Utilizing multi-dimensional features can accurately predict the probability information of the user selecting the candidate material, providing a data foundation for accurately determining search results.

[0140] Among them, the probability information of users selecting candidate materials is the probability information of the candidate material being selected after it is output as a search result. It is used to measure the likelihood of the candidate material being selected by the user.

[0141] For example, a user's selection behavior can be clicking on a material, and the selection probability prediction model can be a click-through rate (CTR) prediction model, with the predicted probability information being the probability information of the user clicking on the candidate material.

[0142] The selected probability prediction model can be obtained by training a deep learning model.

[0143] For example, the probabilistic prediction model can be a deep learning model that combines shallow and deep models, such as DeepFM, Probabilistic Neural Network (PNN), NFM, and Wide&Deep. By combining shallow models (such as the FM layer of DeepFM) and deep models (such as the Deep layer of DeepFM), the relationship between higher-order behavioral representations and material selection can be explored, so that new materials can also get a fairer exposure opportunity in the lead-up scenario.

[0144] For example, when training a deep learning model to obtain a trained selection probability prediction model, a large amount of historical data from the 3D design scenario can be collected, including user design-related asset data, user design behavior logs, material search log data, and material library data. An offline training set and validation set are constructed based on this large amount of historical data. The training and validation sets include training samples and their corresponding annotation information. Each training sample includes the user's long-term preference features, immediate preference features, immediate search features, and material features of a material recommended to the user. The annotation information of the training sample indicates whether the material has been selected by the user. For example, the annotation information can be 0 or 1, where 0 indicates that the material has not been selected by the user, and 1 indicates that the material has been selected by the user.

[0145] When constructing training samples, we can obtain the user's long-term preference features, immediate preference features, and immediate search features based on a single material search behavior, and also obtain the material features of the materials recommended to the user in this search. For each recommended material in this search behavior, the user's long-term preference features, immediate preference features, immediate search features, and the material features of that material can be combined to form a training sample. Samples where the user selects a material are designated as positive samples, and samples where the user does not select a material are designated as negative samples, thus determining the standard information of the training sample. In this way, a large number of labeled training samples can be constructed.

[0146] The deep learning model is trained based on the constructed training samples, so that the deep learning model can learn the relationship between the input features and the user's probability of selecting materials, and obtain a trained selection probability prediction model.

[0147] For example, when training a model, AUC (Area Under Curve) or F1-Score can be used as loss functions.

[0148] Step S308: Output the search results based on the probability information of the user selecting each candidate material.

[0149] After determining the probability information of the user selecting each candidate material, the candidate material with the higher probability of selection is selected as the search result and the search result is output.

[0150] Optionally, based on the probability information of the user selecting each candidate material, information on candidate materials whose probability information is greater than or equal to a probability threshold is used as search results and recommended to the user. The probability threshold can be set and adjusted according to the actual application scenario, and is not specifically limited here.

[0151] Optionally, the candidate materials are sorted based on the probability information of the user selecting each candidate material. A specified number of candidate materials are then presented to the user as search results based on the sorting results. The specified number can be set and adjusted according to the actual application scenario; no specific limitation is made here.

[0152] In addition, when outputting search results, the search results can be displayed in pages according to the offset set by the user, where the offset is the number of materials displayed on each page.

[0153] In this embodiment, long-term user preference features are generated based on the user's design behavior over a longer historical period. Considering that the user's interest in materials changes with search intent and time, real-time user preference features are generated based on the data of the user's current design target solution. Combining the user's long-term preference features and real-time preference features can better characterize the user's preference representation in different search intents and time ranges. Furthermore, by combining the real-time search features determined based on search conditions, the set of materials exposed to the user in actual application scenarios is closer to the user's current search intent.

[0154] For example, after a user (designer) completes a design for a large apartment in a light luxury style, they then begin designing a design for a small apartment in a country style. The user's long-term preferences remain consistent throughout the design process of both designs. However, their immediate preference characteristics are recalculated based on the model selection sequence in the current target design and the information (Meta information) of the design itself. Therefore, when the user's design intent changes, the material search method provided in this application can quickly characterize the design behavior after the change, forming different immediate preference characteristics to better represent the user's current search intent and perform reasonable material search and recommendation.

[0155] In this embodiment, when searching for materials based on user requests, the system acquires the user's long-term preference characteristics for solution design. Explicit user filtering is performed based on search criteria to find candidate materials that match the search conditions. Additionally, the server generates query conditions based on long-term preference characteristics and search criteria to find candidate materials that match the user's long-term preferences, ensuring the search results include materials more closely aligned with those preferences. Furthermore, the system acquires the material characteristics of each candidate material and generates the user's immediate search characteristics based on the search and query conditions to uncover more dimensions of features representing the user's current search intent. The system determines the user's immediate preference characteristics when designing the target solution based on the target solution data. Based on the user's long-term preference characteristics, immediate preference characteristics, immediate search characteristics, and the material characteristics of each candidate material, a selection probability prediction model is used to predict the probability of the user selecting each candidate material, enabling more accurate predictions. Furthermore, based on the probability of the user selecting each candidate material, candidate materials with higher selection probabilities are recommended to the user, making the search results closer to the user's search intent and improving the accuracy of material searches. This reduces the time spent by designers in finding the required materials and increases efficiency.

[0156] Figure 4 A flowchart of a material search method is provided for another exemplary embodiment. The execution entity in this embodiment may be the server mentioned above, such as... Figure 4 As shown, the specific steps of this method are as follows:

[0157] Step S401: Receive the material search request triggered by the user during the design of the target solution, obtain the search conditions, and obtain the user's design behavior representation information and the user's material selection timing behavior representation information when designing the target solution.

[0158] The material search request includes search criteria, user information, and information about the target solution that the user is currently designing.

[0159] Upon receiving a physical search request, the request can be parsed to obtain the search criteria, user information, and target solution information.

[0160] Search criteria are the filter conditions set by users when searching for materials through an interactive page. They are the filter conditions used by users to select materials in the material library.

[0161] User information can include unique identifiers such as user ID, user account, and mobile phone number. Based on this information, the user's long-term preference characteristics for solution design can be retrieved from offline-generated long-term user preference features, serving as a representation of the user's design behavior.

[0162] In this embodiment, the specific implementation method for obtaining the user's design behavior representation information (i.e., the user's long-term preference characteristics in scheme design) is consistent with the method embodiment described above, and will not be repeated here. The target scheme information can be the target scheme's identification information, such as scheme ID, scheme name, or other information that can uniquely identify the target scheme. Based on the target scheme information, the data of the target scheme can be obtained. Based on the target scheme data, the material selection time sequence behavior representation information of the user when designing the target scheme can be determined.

[0163] The information representing the user's material selection sequence behavior when designing the target solution includes at least the user's attribute preference characteristics for each category, statistically derived from the selection sequence of materials already selected in the target solution. The specific implementation of the user's attribute preference characteristics for each category, statistically derived from the selection sequence of materials already selected in the target solution, is consistent with the above method embodiments and will not be repeated here.

[0164] Optionally, the material selection time-series behavior representation information when a user designs a target solution may also include solution characteristic information such as the solution area, number of spaces, creation method, and creation environment. In this case, the material selection time-series behavior representation information when a user designs a target solution is the real-time preference feature of the user when designing the target solution, determined based on the data of the target solution in the above embodiments. The specific acquisition method is the same as in the above embodiments and will not be repeated here.

[0165] Step S402: After finding candidate materials that meet the search criteria, recommend materials based on the user's design behavior representation information and the user's material selection time sequence behavior representation information when designing the target solution, so as to recommend candidate materials that match the user's design behavior preferences.

[0166] After obtaining the search criteria, a material search can be performed based on the search criteria to recall materials that meet the search criteria from the material library as candidate materials.

[0167] After finding candidate materials that match the search criteria, the system recommends materials based on the user's design behavior and the timing of material selection behavior when designing the target solution. This further filters the candidate materials to obtain those that better match the user's design behavior preferences. The filtered candidate materials that match the user's design behavior preferences are then recommended to the user, making the search results more consistent with the user's design behavior preferences and closer to the user's search intent. This improves the accuracy of material search and reduces the time and efficiency for designers to find the materials they need.

[0168] For example, the probability of a user selecting each candidate material can be predicted based on the user's design behavior characteristics, the user's material selection sequence behavior characteristics when designing the target solution, and the material characteristics of the candidate materials. Based on the probability of the user selecting each candidate material, material recommendations are made to suggest candidate materials that match the user's design behavior preferences.

[0169] In an optional embodiment, the user's real-time search characteristics can also be generated based on the search conditions. In step S402 above, material recommendations can be made based on the user's design behavior representation information, the user's material selection time sequence behavior representation information when designing the target solution, and the user's real-time search characteristics. Candidate materials that better match the user's design behavior preferences and the user's current search intent are selected from the candidate materials to recommend candidate materials that better match the user's design behavior preferences and the user's current search intent, thereby improving the accuracy of material search and reducing the time spent by designer users in finding the required materials, thus improving efficiency.

[0170] The method for generating real-time search features of users based on search criteria can be implemented using the same method described in the above examples, and will not be elaborated further here.

[0171] Optionally, the server can also generate query conditions based on long-term preference characteristics and search conditions, and search for candidate materials that meet the query conditions. For the specific implementation method, please refer to the specific implementation method of step S303 above, which will not be repeated here. The candidate materials searched based on the search conditions and query conditions are all treated as candidate materials and processed in step S402.

[0172] Furthermore, when generating a user's real-time search features, the user's real-time search features can be generated based on the search conditions and query conditions. For the specific implementation method, please refer to the specific implementation method of step S305 above, which will not be repeated here.

[0173] Furthermore, in step S402 above, the probability of the user selecting each candidate material can be predicted based on the user's design behavior representation information, the user's material selection time sequence behavior representation information when designing the target solution, the user's real-time search characteristics, and the material characteristics of the candidate materials. For the specific implementation, please refer to step S307 above, which will not be repeated here. Further, based on the probability information of the user selecting each candidate material, material recommendations are made to recommend candidate materials that match the user's design behavior preferences. For the specific implementation, please refer to step S308 above, which will not be repeated here.

[0174] Figure 5 This is a schematic diagram of a material search apparatus provided in an exemplary embodiment of this application. The apparatus provided in this embodiment is applied to a server where a 3D design tool is located, such as... Figure 5 As shown, the material search device 50 includes: a data acquisition module 51, a search module 52, a feature acquisition module 53, a selection probability prediction module 54, and a search result processing module 55.

[0175] The data acquisition module 51 is used to receive material search requests triggered by users during the design of target solutions, and to acquire search conditions, long-term preference characteristics of users in solution design, and data of target solutions.

[0176] Search module 52 is used to search for candidate materials that meet the search criteria.

[0177] The feature acquisition module 53 is used to acquire the material features of each candidate material; generate the user's real-time search features based on the search conditions; and determine the user's real-time preference features when designing the target solution based on the target solution data.

[0178] The probability prediction module 54 is used to predict the probability information of the user selecting each candidate material based on the user's long-term preference characteristics, immediate preference characteristics, immediate search characteristics and material characteristics of each candidate material.

[0179] The search result processing module 55 is used to output search results based on the probability information of the user selecting each candidate material.

[0180] The apparatus provided in this embodiment can be specifically used to perform the above-described... Figure 2 The specific functions and technical effects of the solutions provided in the corresponding method embodiments will not be elaborated here.

[0181] In one optional embodiment, before acquiring the material characteristics of each candidate material, the search module is further configured to:

[0182] Query criteria are generated based on long-term preference characteristics and search conditions, and candidate materials that meet the query criteria are searched. The query criteria are used to search for materials that match the user's long-term preferences.

[0183] In one optional embodiment, the feature acquisition module is further configured to:

[0184] Obtain design-related asset data and material search log data of users within a preset historical period; based on the design-related asset data and material search log data of each user within the preset historical period, determine the long-term preference characteristics of each user, including asset distribution characteristics and category attribute preference characteristics.

[0185] The asset distribution characteristics include at least one of the following: total number of schemes, average number of spaces per scheme, average number of saves per scheme, average number of renders per scheme, proportion of selected commodity materials, and proportion of selected public warehouse materials; the category attribute preference characteristics include at least one of the following: the number of materials selected by users for each attribute under each category, and the material attributes preferred by users.

[0186] In one optional embodiment, when generating query conditions based on the user's long-term preference characteristics and search criteria, the search module is further configured to:

[0187] Determine the target category of the candidate materials found based on the search criteria; generate query criteria based on category attribute preference features, and use the query criteria to search for candidate materials with user-preferred material attributes under the target category.

[0188] In an optional embodiment, when generating real-time search features for a user based on search criteria, the feature acquisition module is further configured to:

[0189] Generate the user's real-time search characteristics based on search criteria and query conditions.

[0190] In one optional embodiment, when generating a user's instant search features based on search criteria and query criteria, the feature acquisition module is further configured to perform at least one of the following processes:

[0191] Based on the preset index dictionary, the condition values ​​of the tag class in the search conditions and query conditions are encoded into corresponding numerical features, and the user's real-time search features include numerical features.

[0192] The search and query conditions are processed by word embedding to generate corresponding word embedding vectors. The user's real-time search features include word embedding vectors.

[0193] In one optional embodiment, when acquiring the material characteristics of each candidate material, the feature acquisition module is further configured to:

[0194] Obtain the label information, category, and attribute information of each candidate material to obtain the material characteristics of each candidate material.

[0195] In one optional embodiment, the user's immediate preference features include the scheme features of the target scheme and the category attribute preference features of the user when designing the target scheme. When determining the user's immediate preference features when designing the target scheme based on the target scheme data, the feature acquisition module is further configured to:

[0196] Based on the data of the target solution, determine the solution characteristics, which include at least one of the following: total area, number of spaces, creation method, and creation environment information; based on the data of the target solution, determine the material selection sequence of the target solution; based on the material selection sequence of the target solution, generate the category attribute preference characteristics of the user when designing the target solution, which include at least one of the following: the quantity of materials for each attribute under each category selected in the target solution, the material attributes preferred by the user in the target solution, and the material attribute selection sequence for each category and attribute.

[0197] In an optional embodiment, when predicting the probability of a user selecting each candidate material based on the user's long-term preference characteristics, immediate preference characteristics, immediate search characteristics, and the material characteristics of each candidate material, the selection probability prediction module is further configured to:

[0198] For each candidate material, the user's long-term preference features, real-time preference features, and real-time search features are concatenated with the material features of the candidate material to obtain concatenated features. The concatenated features are then subjected to dimensionality reduction processing, and the dimensionality-reduced features are input into a trained selection probability prediction model. The selection probability prediction model then predicts the probability information of the user selecting the candidate material.

[0199] The device provided in this embodiment can be used to execute the solution provided in any of the above method embodiments. The specific functions and technical effects that can be achieved will not be described in detail here.

[0200] Figure 6 This is a schematic diagram of a material search apparatus provided in another exemplary embodiment of this application. The apparatus provided in this embodiment is applied to a server where a 3D design tool is located, such as... Figure 6 As shown, the material search device 60 includes a behavior representation module 61 and a material search module 62.

[0201] The behavior representation module 61 is used to receive material search requests triggered by the user during the design of the target solution, obtain search conditions, as well as the user's design behavior representation information and the material selection sequence behavior representation information when the user designs the target solution.

[0202] The material search module 62 is used to recommend candidate materials that match the user's design behavior preferences based on the user's design behavior representation information and the material selection time sequence behavior representation information when the user designs the target scheme, after finding candidate materials that meet the search conditions.

[0203] The apparatus provided in this embodiment can be specifically used to perform the above-described... Figure 4 The specific functions and technical effects of the solutions provided by the corresponding method embodiments and optional embodiments will not be elaborated here.

[0204] Figure 7 This is a schematic diagram of the structure of an electronic device provided in an example embodiment of this application. Figure 7 As shown, the electronic device 70 includes a processor 701 and a memory 702 communicatively connected to the processor 701, the memory 702 storing computer execution instructions.

[0205] The processor executes computer execution instructions stored in the memory to implement the solution provided in any of the above method embodiments. The specific functions and technical effects that can be achieved will not be elaborated here.

[0206] This application also provides a computer-readable storage medium storing computer-executable instructions. When executed by a processor, the computer-executable instructions are used to implement the solution provided in any of the above method embodiments. The specific functions and technical effects to be achieved are not described here.

[0207] This application also provides a computer program product, which includes a computer program stored in a readable storage medium. At least one processor of the electronic device can read the computer program from the readable storage medium. The at least one processor executes the computer program to cause the electronic device to perform the solution provided in any of the above method embodiments. The specific functions and technical effects that can be achieved are not described here.

[0208] Furthermore, in some of the processes described in the above embodiments and accompanying drawings, multiple operations appear in a specific order. However, it should be clearly understood that these operations may not be executed in the order they appear herein, or may be executed in parallel. The sequence numbers are merely used to distinguish different operations, and the sequence number itself does not represent any execution order. Additionally, these processes may include more or fewer operations, and these operations may be executed sequentially or in parallel. It should be noted that the descriptions such as "first," "second," etc., in this document are used to distinguish different messages, devices, modules, etc., and do not represent a sequential order, nor do they limit "first" and "second" to different types. "Multiple" means two or more, unless otherwise explicitly specified.

[0209] Other embodiments of this application will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of this application that follow the general principles of this application and include common knowledge or customary techniques in the art not disclosed herein. The specification and examples are to be considered exemplary only, and the true scope and spirit of this application are indicated by the following claims.

[0210] It should be understood that this application is not limited to the precise structure described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of this application is limited only by the appended claims.

Claims

1. A method of material search, characterized by, The method comprises the following steps: receiving a material search request triggered by a user during the design of a target scheme, obtaining search conditions, long-term preference characteristics of the user for scheme design, and data of the target scheme; searching for candidate materials that meet the search conditions, and obtaining material characteristics of each candidate material; generating instant search characteristics of the user according to the search conditions; determining instant preference characteristics of the user for designing the target scheme according to the data of the target scheme; the instant preference characteristics of the user include scheme characteristics of the target scheme and category attribute preference characteristics of the user for designing the target scheme; predicting probability information of the user for selecting each candidate material according to the long-term preference characteristics, instant preference characteristics, instant search characteristics of the user, and material characteristics of each candidate material; outputting a search result according to the probability information of the user for selecting each candidate material; determining the instant preference characteristics of the user for designing the target scheme according to the data of the target scheme, including determining scheme characteristics of the target scheme according to the data of the target scheme, and the scheme characteristics including at least one of the following: total area, space number, creation mode, and creation environment information; determining a material selection sequence of the target scheme according to the data of the target scheme, and generating category attribute preference characteristics of the user for designing the target scheme according to the material selection sequence of the target scheme, and the category attribute preference characteristics of the user for designing the target scheme including at least one of the following: the number of materials of each attribute in each category that have been selected in the target scheme, material attributes preferred by the user in the target scheme, and material attribute selection sequences of each attribute in each category.

2. The method of claim 1, wherein, Before obtaining the material characteristics of each candidate material, the method further comprises the following steps: generating a query condition according to the long-term preference characteristics and the search conditions, and searching for candidate materials that meet the query condition, wherein the query condition is used to search for materials that meet the long-term preference of the user.

3. The method of claim 2, wherein, The method further comprises the following steps: obtaining design-related asset data and material search log data of the user within a preset historical period; determining long-term preference characteristics of each user according to the design-related asset data and material search log data of each user within a preset historical period, and the long-term preference characteristics including asset distribution characteristics and category attribute preference characteristics; the asset distribution characteristics include at least one of the following: total number of schemes, average space number of schemes, average number of times of saving schemes, average number of times of rendering schemes, proportion of selected commodity materials, and proportion of selected public library materials; the category attribute preference characteristics include at least one of the following: selected number of materials of each attribute in each category by the user, and material attributes preferred by the user.

4. The method of claim 3, wherein, The method further comprises the following steps: determining a target category in which the candidate materials searched according to the search conditions are located; generating a query condition according to the category attribute preference characteristics, and the query condition is used to search for candidate materials with the material attributes preferred by the user in the target category.

5. The method of claim 2, wherein, The generating the instant search feature of the user according to the search condition comprises: The instant search feature of the user is generated according to the search condition and the query condition.

6. The method of claim 5, wherein, The generating the instant search feature of the user according to the search condition and the query condition comprises at least one of the following: According to a preset index dictionary, the condition value of the label type in the search condition and the query condition is encoded into a corresponding numerical feature, and the instant search feature of the user comprises the numerical feature; The search condition and the query condition are subjected to word embedding processing to generate a corresponding word embedding vector, and the instant search feature of the user comprises the word embedding vector.

7. The method of claim 1, wherein, The obtaining the material feature of each candidate material comprises: The label information, the category to which the candidate material belongs and the attribute information of each candidate material are obtained to obtain the material feature of each candidate material.

8. The method according to any one of claims 1-7, characterized in that, The predicting the probability information of the user selecting each candidate material according to the long-term preference feature, the instant preference feature, the instant search feature of the user and the material feature of each candidate material comprises: For each candidate material, the long-term preference feature, the instant preference feature, the instant search feature of the user and the material feature of the candidate material are spliced to obtain spliced features; The spliced features are subjected to dimension reduction processing, and the features after the dimension reduction processing are input into a trained selection probability estimation model to predict the probability information of the user selecting the candidate material through the selection probability estimation model.

9. A method of material search, characterized by, It comprises: A material search request triggered by a user in the process of designing a target scheme is received, a search condition is obtained, and design behavior representation information of the user and material selection time sequence behavior representation information of the user when designing the target scheme are obtained, the material selection time sequence behavior representation information of the user when designing the target scheme being instant preference feature of the user when designing the target scheme determined according to data of the target scheme; The instant preference feature of the user comprises scheme feature of the target scheme and category attribute preference feature of the user when designing the target scheme; After candidate materials meeting the search condition are searched, material recommendation is performed according to the design behavior representation information of the user and the material selection time sequence behavior representation information of the user when designing the target scheme, so as to recommend candidate materials meeting the design behavior preference of the user to the user; The determining the instant preference feature of the user when designing the target scheme according to the data of the target scheme comprises: determining scheme feature of the target scheme according to the data of the target scheme, the scheme feature comprising at least one of the following: total area, space number, creation mode and creation environment information; The material selection sequence of the target scheme is determined according to the data of the target scheme, and the category attribute preference feature of the user when designing the target scheme is generated according to the material selection sequence of the target scheme, and the category attribute preference feature of the user when designing the target scheme includes at least one of the following: the number of materials of each attribute under each category in the target scheme, the material attribute preferred by the user in the target scheme, and the material attribute selection sequence of each attribute under each category.

10. An apparatus for material search, characterized by, Comprise: The data acquisition module is used for receiving the material search request triggered by the user during the design of the target scheme, acquiring the search condition, the long-term preference feature of the user during the scheme design, and the data of the target scheme; The search module is used for searching the candidate materials meeting the search condition; The feature acquisition module is used for acquiring the material feature of each candidate material; The instant search feature of the user is generated according to the search condition; And the instant preference feature of the user when designing the target scheme is determined according to the data of the target scheme; The instant preference feature of the user includes the scheme feature of the target scheme and the category attribute preference feature of the user when designing the target scheme; The selection probability prediction module is used for predicting the probability information of the user selecting each candidate material according to the long-term preference feature, the instant preference feature, the instant search feature of the user and the material feature of each candidate material; The search result processing module is used for outputting the search result according to the probability information of the user selecting each candidate material; The feature acquisition module is specifically used for determining the scheme feature of the target scheme according to the data of the target scheme, and the scheme feature includes at least one of the following: total area, space number, creation mode and creation environment information; The material selection sequence of the target scheme is determined according to the data of the target scheme, and the category attribute preference feature of the user when designing the target scheme is generated according to the material selection sequence of the target scheme, and the category attribute preference feature of the user when designing the target scheme includes at least one of the following: the number of materials of each attribute under each category in the target scheme, the material attribute preferred by the user in the target scheme, and the material attribute selection sequence of each attribute under each category.

11. An apparatus for material search, characterized by Comprise: The behavior characterization module is used for receiving the material search request triggered by the user during the design of the target scheme, acquiring the search condition, and the design behavior characterization information of the user and the material selection time sequence behavior characterization information of the user when designing the target scheme, and the material selection time sequence behavior characterization information of the user when designing the target scheme is the instant preference feature of the user when designing the target scheme determined according to the data of the target scheme; The instant preference feature of the user includes the scheme feature of the target scheme and the category attribute preference feature of the user when designing the target scheme; The material searching module is configured to, after searching for the candidate material meeting the search condition, recommend the candidate material meeting the design behavior preference of the user to the user according to the design behavior characteristic information of the user and the material selection time sequence characteristic information when the user designs the target scheme. The determining, according to the data of the target scheme, of the instant preference feature of the user when designing the target scheme comprises: determining, according to the data of the target scheme, a scheme feature of the target scheme, the scheme feature comprising at least one of the following: total area, space number, creation mode, and creation environment information. The determining, according to the data of the target scheme, of the material selection sequence of the target scheme, and the generating of the category attribute preference feature of the user when designing the target scheme according to the material selection sequence of the target scheme, the category attribute preference feature of the user when designing the target scheme comprising at least one of the following: the number of materials of each attribute under each category that has been selected in the target scheme, the material attribute preferred by the user in the target scheme, and the material attribute selection sequence of each category attribute under each category.

12. An electronic device, comprising: The method comprises: a processor, and a memory connected in communication with the processor; the memory stores computer-executed instructions; the processor executes the computer-executed instructions stored in the memory to implement the method according to any one of claims 1-9.

13. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-executed instructions, and the computer-executed instructions are executed by the processor to implement the method according to any one of claims 1-9.

Images