Commodity display method and device, electronic equipment and storage medium
By constructing an agent-based product preference identifier, utilizing a dual-tower matching model and a multi-task learning mechanism, and combining user features and multimodal features, the problem of inaccurate game account recommendations in existing technologies is solved, thereby improving the user experience of online trading platforms.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- YIDIAN LINGXI INFORMATION TECHNOLOGY (GUANGZHOU) CO LTD
- Filing Date
- 2026-01-21
- Publication Date
- 2026-06-05
Smart Images
Figure CN122155805A_ABST
Abstract
Description
Technical Field
[0001] This disclosure relates to the technical field of online transactions, and more specifically, to a product display method, apparatus, electronic device, and storage medium. Background Technology
[0002] With the widespread adoption of online transactions, users are buying and selling online more frequently. Currently, to facilitate the trading of virtual goods, sellers can list game items for sale on online trading platforms, and buyers can purchase the game items they need. However, the highly heterogeneous attributes of game accounts make it difficult to recommend highly accurate and preferred game accounts to buyers, resulting in a poor user experience. Summary of the Invention
[0003] One objective of this disclosure is to provide a new technical solution for product display.
[0004] According to a first aspect of this disclosure, a product display method is provided, the method comprising: Obtain the buyer's browsing history information; wherein, the browsing history information includes the buyer's browsing history products, attribute information representing the product attributes of the historical products and the subordinate attributes of the subordinate assets configured for the historical products, and the buyer's interaction behavior identifiers with the historical products; The first intelligent agent is invoked to construct a product preference identifier for the buyer user based on the attribute information and the interaction behavior identifier; The second intelligent agent is invoked to determine candidate products that match the product preference identifier from a pre-set product library; wherein, the candidate products are used to be displayed preferentially when the buyer user browses.
[0005] Optionally, the step of invoking the first intelligent agent to construct a product preference identifier for the buyer user based on the attribute information and the interaction behavior identifier includes: The first intelligent agent is invoked to construct the shared features of the historical products and the product-side features of the shared features under the target interaction behavior based on the attribute information. The user-side features are determined based on the interaction behavior identifiers; The product-side features and the user-side features are concatenated to obtain fused features; The fused features are semantically transformed to construct a first semantic feature that represents the buyer user's interaction behavior, and this feature serves as an identifier of the buyer user's product preferences.
[0006] Optionally, the method further includes: Invoke the first intelligent agent to obtain the user characteristics of the buyer user; Determining the user-side features based on the interaction behavior identifier includes: The user-side features are determined based on the user characteristics and the interaction behavior identifiers.
[0007] Optionally, the method further includes: Acquire first user behavior data; wherein, the first user behavior data includes the first user identifier of the first user, the first product identifier of the first product, and the target interaction behavior identifier of the first user's first interaction behavior with the first product; Based on the target behavior relationship chain reflected by the target interaction behavior identifier, determine the sample type of the first user behavior data and construct training sample pairs; The first agent is trained using the training sample pairs to obtain a trained first agent.
[0008] Optionally, the method further includes: Invoke a third-party intelligent agent to obtain the seller user's input information regarding the historical products; The input information is subjected to feature transformation processing to obtain multimodal features; The step of semantically transforming the fused features to construct the first semantic feature includes: The multimodal features and the fused features are concatenated to obtain the first semantic feature representing the seller's user interaction behavior.
[0009] Optionally, the step of invoking the second intelligent agent to determine candidate products matching the product preference identifier from a pre-set product library includes: The second intelligent agent is invoked to generate a product feature identifier that matches the product characteristics expected by the buyer user, based on the product preference identifier. In a pre-set product library, candidate products corresponding to the product feature identifier are identified.
[0010] Optionally, the method further includes: Acquire second user behavior data; wherein, the second user behavior data includes the second product in which the second user interacts, the second product identifier of the second product, and the interaction time with the second product; Determine the semantic feature encoding of the product mapped to the second product identifier; Based on the interaction time, the multiple semantic features of the products are divided into input samples and output samples; The second agent is trained using the input samples and the output samples to obtain the trained generative model.
[0011] According to a second aspect of this disclosure, a product display device is also provided, the device comprising: The acquisition module is used to acquire the buyer's historical browsing information; wherein, the historical browsing information includes the historical products browsed by the buyer, the product attributes representing the historical products and the attribute information of the subordinate assets configured for the historical products, and the identifiers of the buyer's interaction behavior with the historical products; The construction module is used to invoke the first intelligent agent to construct a product preference identifier for the buyer user based on the attribute information and the interaction behavior identifier; The determination module is used to invoke the second intelligent agent to determine candidate products that match the product preference identifier in a preset product library; wherein, the candidate products are used to be displayed preferentially when the buyer user browses.
[0012] According to a third aspect of this disclosure, a computer system is also provided, the computer system including a processor, which implements the product display method of the first aspect when the processor executes program instructions or code.
[0013] For example, the computer system also includes a memory for storing program instructions or code.
[0014] According to a fourth aspect of this disclosure, a computer-readable storage medium is also provided, wherein a computer program is stored therein, wherein the computer program is configured to execute the above-described product display method at runtime.
[0015] According to a fifth aspect of this disclosure, a computer program product is also provided, comprising a computer program that, when executed, causes a computer to perform the steps of the above-described product display method.
[0016] According to a sixth aspect of this disclosure, an electronic device is also provided, including a memory and a processor, wherein the memory stores a computer program and the processor is configured to execute the product display method described above through the computer program.
[0017] One beneficial effect of this disclosure is that the product display method provided by the present invention can obtain interaction behavior identifiers representing user interaction behavior from the buyer's historical browsing information, invoke a first intelligent agent, and construct the buyer's product preference identifier based on the attribute information of historical products and the interaction behavior identifiers. Then, a second intelligent agent is invoked to filter out candidate products in the product library that match the product preference identifiers, and recommend them to the buyer. This achieves the goal of determining the buyer's key focus products through the buyer's interaction behavior identifiers, and recommending similar candidate products, effectively improving the accuracy of product recommendations and thus enhancing the buyer's online shopping experience.
[0018] Other features and advantages of the embodiments of this disclosure will become clear from the following detailed description of exemplary embodiments with reference to the accompanying drawings. Attached Figure Description
[0019] The accompanying drawings, which are incorporated in and form a part of this specification, illustrate embodiments of the present disclosure and, together with their description, serve to explain the principles of the embodiments of the present disclosure.
[0020] Figure 1 A schematic diagram of the hardware structure of an electronic device that can be used to implement a product display method according to embodiments of the present disclosure is shown; Figure 2 A flowchart illustrating a product display method according to some embodiments is shown; Figure 3 A schematic diagram of the structure of a first intelligent agent according to some embodiments is shown; Figure 4 A schematic diagram illustrating the training of a first agent according to some embodiments is shown; Figure 5 A schematic diagram illustrating the training of a first agent according to some other embodiments is shown; Figure 6 A schematic diagram of the structure of a second intelligent agent according to some embodiments is shown; Figure 7 A schematic diagram of the structure of a product display device according to some embodiments is shown; Figure 8 A schematic diagram of the hardware structure of an electronic device according to some embodiments is shown. Detailed Implementation
[0021] Various exemplary embodiments of this specification will now be described in detail with reference to the accompanying drawings.
[0022] The following description of at least one exemplary embodiment is merely illustrative and is in no way intended to limit the embodiments of this specification or their application or use.
[0023] It should be noted that similar labels and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be discussed further in subsequent figures.
[0024] It should be noted that all actions involving the acquisition of signals, information, or data in this embodiment are carried out in compliance with the relevant data protection laws and regulations of the country where the location is situated, and with authorization from the owner of the relevant equipment.
[0025] This disclosure provides a novel product display scheme that acquires historical browsing information from online trading platforms and then uses this information to determine product preference identifiers that indicate the products the buyer is particularly interested in. Candidate products matching these preference identifiers are identified from all products listed on the online trading platform, and these candidate products are prioritized for display to the buyer, effectively improving the accuracy of product recommendations and thus enhancing the online shopping experience for the buyer.
[0026] Figure 1 A schematic diagram of the hardware structure of an electronic device that can be used to implement a product display according to an embodiment of the present disclosure is shown.
[0027] The electronic device 1000 is a device capable of running computer programs. These programs can be local applications installed on the electronic device, or web applications, lightweight applications, or mini-programs, etc., without limitation. The electronic device 1000 can be a mobile phone, tablet computer, PC, etc., without limitation.
[0028] like Figure 1 As shown, the electronic device 1000 may include a processor 1101, a memory 1102, an interface device 1103, a communication device 1104, an output device 1105, an input device 1106, etc. Figure 1 The hardware configuration shown is illustrative only and is not intended to limit this disclosure, its application, or its use.
[0029] The processor 1101 executes computer programs, which can be written using instruction sets of architectures such as x86, Arm, RISC, MIPS, and SSE. The memory 1102 includes, for example, ROM (Read-Only Memory), RAM (Random Access Memory), and non-volatile memory such as a hard disk. The interface device 1103 includes, for example, a USB interface, a network cable interface, and a headphone jack. The communication device 1104 is capable of wired or wireless communication. The communication device 1104 may include at least one short-range communication module, such as any module for short-range wireless communication based on short-range wireless communication protocols such as Hilink, WiFi (IEEE 802.11), Mesh, Bluetooth, ZigBee, Thread, Z-Wave, NFC, UWB, and LiFi. The communication device 1104 may also include a long-range communication module, such as any module for WLAN, GPRS, or 2G / 3G / 4G / 5G long-range communication. The output device 1105 may include, for example, an LCD screen or touch screen, and a speaker. Input device 1106 may include, for example, a touch screen, a keyboard, a microphone, various sensors, etc.
[0030] In this embodiment, the memory 1102 of the electronic device 1000 is used to store a computer program that controls the processor 1101 to operate in order to execute a product display method according to any embodiment of this disclosure.
[0031] Next, as Figure 1 Taking an electronic device 1000 as an example, we will illustrate various embodiments of the product display method.
[0032] <First Embodiment> Figure 2 A product display method according to some embodiments is shown, which may include the following steps S210 to S230: Step S210: Obtain the buyer's historical browsing information; wherein, the historical browsing information includes the historical products browsed by the buyer, the product attributes representing the historical products and the attribute information of the subordinate assets configured for the historical products, and the identifiers of the buyer's interaction behavior with the historical products.
[0033] In this embodiment, the product can be a virtual product. The product attributes of the virtual product can be the product attributes that the product itself has. For example, if the virtual product is a game account, the product attributes of the virtual product can include the game account ID, the username of the game account, and the account creation time, etc.
[0034] In this embodiment, the goods here can be virtual goods with at least one subordinate asset. For example, the virtual goods are game accounts, and the subordinate asset can be the game assets possessed by the game account, including virtual characters. Furthermore, the subordinate attributes of the subordinate asset can include the character attributes of the virtual character, the virtual items held by the virtual character, and the virtual vehicles held by the virtual character, etc.
[0035] In some examples, the historical items represent game accounts viewed by the buyer during a specific time period. The game account configuration information can include the character account information and attribute information of the characters associated with that account. The character account information can include character class, level, or gender, etc., and this information can be displayed on the game's character login screen. The character attribute information can include character attributes, virtual items held by the character, and virtual vehicles held by the character, etc., and this attribute information can be displayed on the character usage screen after the character login screen.
[0036] In some examples, a buyer's interaction with a historical product may include clicking "favorite and place an order," clicking "favorite but not placing an order," clicking "place an order but not favorite," and clicking "not favorite and not place an order." The interaction identifier can be a sequence of actions. For example, a historical product's action sequence [click:1,fav:1,order:1] indicates that the user clicked "favorite and placed an order" for that historical product. Another example is a historical product's action sequence [click:1,fav:1,order:0], which indicates that the user clicked "favorite" but did not place an order.
[0037] Step S220: Invoke the first intelligent agent to construct product preference identifiers for buyer users based on attribute information and interaction behavior identifiers.
[0038] In this embodiment, the first intelligent agent can be an artificial intelligence network model, which has the functions of information extraction and semantic generation. The first intelligent agent can be a general artificial intelligence model or an optimized or trained artificial intelligence model, and there is no limitation here.
[0039] In some examples, the product preference identifier constructed by the first agent based on attribute information and interaction behavior identifiers can be a collaborative vector representing the user's click preference, collection preference, and order placement preference.
[0040] In some examples, step S220 may be: when a buyer triggers a product recommendation control, a first intelligent agent is invoked to construct a product preference identifier for the buyer based on attribute information and interaction behavior identifiers.
[0041] In this embodiment, the product recommendation control can be set in a specific area of the product browsing interface of the online trading platform. After the buyer user triggers the product recommendation control, he / she can construct a product preference identifier to realize the target product item of the candidate products in the subsequent product browsing interface.
[0042] In some examples, step S220 may be: when a buyer logs back into the online trading platform after a set time interval and / or a set number of times, the first intelligent agent is invoked to construct a product preference identifier for the buyer based on attribute information and interaction behavior identifiers.
[0043] In this embodiment, the set duration and set number of times can be set manually and are not limited here. After the buyer user logs back into the online trading platform after the set duration and / or set number of times, the product items displayed on the product browsing interface of the online trading platform are refreshed, which can build product preference identifiers to realize the target product items of candidate products in the subsequent product browsing interface.
[0044] Step S230: Invoke the second intelligent agent to determine candidate products that match the product preference identifier in the preset product library; wherein, the candidate products are used to be displayed preferentially when the buyer user browses.
[0045] In this embodiment, the second intelligent agent can be an artificial intelligence network model with semantic mapping capabilities. The second intelligent agent can be a general-purpose artificial intelligence model or an optimized or trained artificial intelligence model; no limitation is made here. The second intelligent agent can also be a first intelligent agent, which can simultaneously possess information extraction, semantic generation, and semantic mapping capabilities.
[0046] In this embodiment, the product database can be a database that stores product information of all virtual products on the online trading platform. All virtual products here have multiple states, which can include listed, delisted, sold, and currently being traded. The product database can store product information of virtual products in a specific state, such as listed.
[0047] In some examples, the priority display could be: when browsing products on an online trading platform, the candidate products corresponding to the target product item displayed to the buyer user on the product browsing interface, where the target product item could be one of the first few product items in the product browsing interface.
[0048] In some examples, step S230 may be: if the preset product library meets the set query conditions, call the second intelligent agent to determine candidate products that match the product preference identifier in the preset product library; wherein, the candidate products are used to be displayed preferentially when the buyer user browses.
[0049] In this embodiment, the query conditions may include the quantity of products in the product database being greater than or equal to a set quantity and the product database being in an update-completed state. The update status of the product database may be that information about other products is being updated into the product database, and the update-completed status of the product database may be that the information updates for other products have been completed.
[0050] The product display method according to the first embodiment of the present invention solves the problem of highly heterogeneous account attributes, making it difficult to recommend highly accurate preferred game accounts to buyers. Based on this method, interaction behavior identifiers representing user interaction behavior are obtained from the buyer's historical browsing information. A first intelligent agent is invoked to construct the buyer's product preference identifier based on the attribute information of historical products and the interaction behavior identifiers. A second intelligent agent is then invoked to filter candidate products in the product library that match the product preference identifier, and recommends them to the buyer. This achieves the goal of determining the buyer's key interest products through the buyer's interaction behavior identifiers, and recommending similar candidate products, effectively improving the accuracy of product recommendations and thus enhancing the online shopping experience for buyers.
[0051] <Second Embodiment> In this embodiment, in order to construct the product preference identifier of the buyer user, the first intelligent agent is invoked to construct the product-side features of the products that the buyer user focuses on and the user-side features that represent the buyer user's interactive behavior. Then, the product preference identifier that represents the buyer user's behavior is constructed through the product-side features and the user-side features.
[0052] In these embodiments, relative to the first embodiment described above, step S220 may include the following steps S310 to S340: Step S310: Invoke the first intelligent agent to construct the shared features of historical products and the product-side features of the shared features under the target interaction behavior based on the attribute information.
[0053] In this embodiment, the first intelligent agent is based on an artificial intelligence network model constructed using a dual-tower matching model (Deep Structured Semantic Model, DSSM) and a multi-task learning mechanism (Multi-gate Mixture-of-Experts, MMoE), such as... Figure 3 As shown, the first intelligent agent can obtain shared embeddings through attribute information such as item ID, item price, game ID, seller ID, and equipment attributes of historical goods, in order to construct shared features. The expression is as follows:
[0054] In the formula, MLP () represents a multilayer perceptron. Emb ( ID This represents the feature embedding of the item ID. Emb ( price ) represents the feature embedding of item price, Emb ( dThis indicates the feature embedding of the game ID. Emb ( d This indicates the feature embedding of the seller ID. Feature embedding that represents equipment attributes.
[0055] In this embodiment, based on a multi-task learning mechanism, the shared features can be obtained as product-side features under the target interaction behavior, i.e., the clicked product-side features. Features of collectible items and the product-side features of the order The product-side features clicked here The expression is as follows:
[0056] In the formula, This indicates the gating parameters set for the click. This indicates the weighting coefficient assigned to each click. ( () represents the k-th expert network, typically a multilayer perceptron, which performs feature transformation on shared features and outputs expert features. This process continues to obtain the features of the collected items. and the product-side features of the order The expression.
[0057] Step S320: Determine user-side features based on interaction behavior identifiers.
[0058] In this embodiment, as Figure 3 As shown, interactive behaviors can correspond to clicks, favorites, and orders. Interactive behavior identifiers can correspond to behavior sequence embeddings. Through interactive behavior identifiers, the user pyramid MLP, i.e., user-side features, can be obtained. The expression is as follows:
[0059] In the formula, Indicates action sequence 1, Let n represent the sequence of actions, where n is a positive integer.
[0060] Step S330: The product-side features and user-side features are concatenated to obtain the fused features.
[0061] In this embodiment, as Figure 3 As shown, concatenating product-side features and user-side features yields a click-specific vector. Vector for collectors , , as a feature of fusion.
[0062] Step S340: Semantically transform the fused features to construct a first semantic feature representing the buyer's user interaction behavior, and use it as an identifier of the buyer's user's product preferences.
[0063] In this set of embodiments, special vectors are used. Vector for collectors , Concatenate into a collaborative vector Using the first semantic feature, the collaborative vector here The expression is as follows:
[0064] In this embodiment, the first agent, constructed based on the Deep Structured Semantic Model (DSSM) and the Multi-gate Mixture-of-Experts (MMoE) mechanism, can obtain product-side features representing the products that buyers are particularly interested in and user-side features representing the interactive behavior of buyers, thereby improving the accuracy of the candidate products obtained subsequently.
[0065] <Third Embodiment> In this embodiment, in order to further improve the accuracy of product recommendations, user-side features are obtained based on interaction behavior identifiers and additionally combined with the user characteristics of the buyer.
[0066] In these embodiments, relative to the second embodiment described above, the method further includes the following step S410 before step S320: Step S410: Invoke the first intelligent agent to obtain the user characteristics of the buyer user.
[0067] In this embodiment, user characteristics may include the buyer's age, gender, etc.
[0068] Based on this, step S320 may include the following step S420: Step S420: Determine user-side characteristics based on user characteristics and interaction behavior identifiers.
[0069] In this embodiment, as Figure 3 As shown, interactive behaviors can correspond to clicks, favorites, and orders; interactive behavior identifiers can correspond to behavior sequence embeddings; and user features can correspond to static feature embeddings. By concatenating the vectors of interactive behavior identifiers and user features, the user-side MLP (Multi-Level Library) can be obtained, i.e., user-side features. The expression is as follows:
[0070] In the formula, Indicates action sequence 1, Let n represent a sequence of lines, where n is a positive integer. Emb ( age ) represents the feature embedding of age. Emb ( gender () represents the feature embedding of gender.
[0071] In this embodiment, by embedding user features, the accuracy of the first intelligent agent is improved, thereby further enhancing the precision of product recommendations.
[0072] <Fourth Embodiment> In this embodiment, in order to further improve the accuracy of the first agent, positive and negative sample pairs are constructed to train the first agent.
[0073] In these embodiments, relative to the first embodiment described above, before step S210, the method further includes the following steps S510 to S530: Step S510: Obtain first user behavior data; wherein, the first user behavior data includes the first user identifier of the first user, the first product identifier of the first product, and the target interaction behavior identifier of the first user's first interaction behavior with the first product.
[0074] In this embodiment, as Figure 4 As shown, the first user behavior database is an online user behavior log. This online user behavior log can include user ID, virtual product ID, and interaction behavior tag. Here, user ID is the first user identifier, virtual product ID is the first product identifier, and target interaction behavior identifier is the interaction behavior tag. Using the first user identifier, first product identifier, and target interaction behavior identifier, a triplet sample (u, a, y) can be constructed, where u represents user ID, a represents first product identifier, and y = (u, a, y). , , () indicates an interactive behavior label.
[0075] Step S520: Based on the target behavior relationship chain reflected by the target interaction behavior identifier, determine the sample type of the first user behavior data and construct training sample pairs.
[0076] In this embodiment, the behavioral relationship chain can include click, no click, click followed by favorite, click without favorite, place order, and favorite but no order. Click, click followed by favorite, and place order can correspond to the sample types of positive samples, while no click, click without favorite, and favorite but no order can correspond to the sample types of negative samples. Therefore, training sample pairs of positive and negative samples can be constructed using the first user behavior data. For example, when the target interaction behavior identifier corresponds to a click, the sample type of the first user behavior data is the sample type of a positive sample.
[0077] Step S530: Train the first agent using training sample pairs to obtain a trained first agent.
[0078] In this embodiment, as Figure 3 As shown, after obtaining the fused features, the matching score and loss value of the first agent can be determined. The matching score may include the click score. Collection Score And order score Click to score and loss value The expression is as follows:
[0079] In the formula, This represents the predicted score for clicks; similarly, the collection score can be obtained. And order score The expression is given. α, β, and γ are the set weighting coefficients. This represents the loss value for clicks. , The loss value of the order.
[0080] In this embodiment, the click loss value is the binary classification cross-entropy, and the click loss value is... The expression is as follows:
[0081] Where N represents the number of samples of the first user behavior data. This represents the true label of the i-th sample, and so on, to obtain... And the loss value of placing an order.
[0082] In this embodiment, after obtaining the loss values for clicks, favorites, and orders, the overall loss value of the first intelligent agent can be obtained. Through this loss value Backpropagation updates the parameters of the first agent.
[0083] <Fifth Embodiment> In this embodiment, to further improve the accuracy of product recommendations, input information such as product screenshots or description text uploaded by seller users is extracted to obtain multimodal features. Based on the fused features and multimodal features, product preference identifiers representing seller user interaction behavior are obtained.
[0084] In these embodiments, relative to the second embodiment described above, before step S330, the method further includes the following steps S610 and S620: Step S610: Invoke the third intelligent agent to obtain the input information of the seller user on historical products.
[0085] In this embodiment, the third intelligent agent can be a general artificial intelligence network model or an optimized or trained artificial intelligence model, without limitation. The third intelligent agent can be a first intelligent agent, which can have the functions of information extraction, semantic generation, semantic mapping, and information extraction and transformation, or it can be a second intelligent agent, which can have the functions of semantic mapping and information extraction and transformation.
[0086] In this embodiment, the input information may be a text description of the seller's historical products sold and / or screenshots from the game.
[0087] Step S620: Perform feature transformation processing on the input information to obtain multimodal features.
[0088] In this embodiment, multimodal features The expression is as follows:
[0089] In the formula, text represents a text description, and image represents an in-game screenshot.
[0090] Based on this, step S330 may include the following step S630: Step S630: The multimodal features and fused features are concatenated to obtain the first semantic features representing the seller's user interaction behavior.
[0091] In this embodiment, the input vector formed by splicing the multimodal features and fusing the features is... x The expression is as follows:
[0092] in, Represents multimodal features, This indicates the fusion feature.
[0093] In this embodiment, the first intelligent agent also includes a semantic generation model (Residual Quantized Variational Autoencoder, RQ-VAE), which can convert the input vector into first semantic features representing the seller's user interaction behavior.
[0094] In some examples, the semantic generation model sets the input vector through hierarchical residual quantization. x The expression is Set the input synthesis vector The latent variables are obtained by the DNN Encoder: Set up m codebooks Each size is K code vector The initial residual is Then quantize layer by layer (i=0 to m) 1): The reconstructed vector is:
[0095] The expression for the first semantic feature, Semantic ID, obtained by the final semantic generation model is as follows:
[0096] In this embodiment, the gradient pass-through estimation of the semantic generation model can be hard quantization used in forward propagation and pass-through gradient used in back propagation, as specifically expressed below:
[0097] In the formula, sg[] represents stop-gradient.
[0098] In this embodiment, as Figure 5 As shown, the loss function of the semantic generation model includes the reconstruction loss function, the codebook loss function, and the total loss function.
[0099] The reconstruction loss function is expressed as follows:
[0100] The codebook loss function is expressed as follows:
[0101] The total loss function is expressed as follows:
[0102] In this embodiment, multimodal features are obtained by extracting input information such as product screenshots or description text uploaded by seller users. Based on the fused features and multimodal features, product preference identifiers representing seller user interaction behavior are obtained, effectively improving the accuracy of product preference identifiers in representing user preferences.
[0103] In some examples, the semantic generation model can be obtained by training on a pool of all game accounts on an online trading platform, extracting collaborative vectors and multimodal vectors, and concatenating them to form a comprehensive vector. Re-encoding Layered quantization is used to find the nearest code vector layer by layer and update the residuals for reconstruction. Then determine the loss value of the semantic generation model. The expression is as follows:
[0104] In the formula, the loss value is... Backpropagation updates the Encoder and Decoder parameters and codebook vector of the semantic generation model. This is to enable model training for semantic generation models.
[0105] <Sixth Embodiment> In this embodiment, in order to filter out candidate products that meet the buyer's preferences, the second intelligent agent generates a product feature identifier that matches the product characteristics expected by the buyer, and then filters candidate products that match the product feature identifier from the product library to prioritize displaying them to the buyer.
[0106] In these embodiments, relative to the second embodiment described above, step S230 may include the following steps S710 and S720: Step S710: Invoke the second intelligent agent to generate a product feature identifier that matches the buyer's desired product characteristics based on the product preference identifier.
[0107] In this embodiment, the second intelligent agent adopts the following... Figure 6 In the Encoder-Decoder architecture shown, the product preference identifier can be represented as an input sequence. Input sequence The expression can be as follows:
[0108] In the formula, [ITEM] represents the separator. Represents a discrete token.
[0109] The encoder expression is as follows:
[0110] The decoder expression is as follows:
[0111] The final generated product feature identifier Pred expression that matches the product characteristics expected by the buyer is as follows:
[0112] Step S720: In the preset product library, determine the candidate products corresponding to the product feature identifiers.
[0113] In this embodiment, by using the aforementioned product feature identifiers, matching candidate products can be obtained from the product database and recommended to buyer users.
[0114] <Seventh Embodiment> In this embodiment, in order to further improve the accuracy of the second agent, input samples and output samples are constructed to train the second agent.
[0115] In these embodiments, relative to the sixth embodiment described above, before step S210, the method further includes the following steps S810 to S840: Step S810: Obtain second user behavior data; wherein, the second user behavior data includes the second product in which the second user interacts, the second product identifier of the second product, and the interaction time with the second product.
[0116] In this embodiment, by using the interaction time of the second products, the second products can be arranged in chronological order to obtain the n most recent second products that the second user has interacted with. That is, the sequence corresponding to the n second products can be [ , ,..., , ].
[0117] Step S820: Determine the semantic feature encoding of the product mapped to the second product identifier.
[0118] In this embodiment, the semantic feature encoding of the mapped product can be determined by the sequence corresponding to the above n second products: [ , ,..., , ].
[0119] Step S830: Based on the interaction time, the semantic features of multiple products are encoded and divided into input samples and output samples.
[0120] In this embodiment, the input sample is represented as [ , ,..., The output sample is represented as =[ , ,..., ].
[0121] In some examples, the semantic feature encoding of the product corresponding to the behavior sequence ABCD is (103,211,56)-(889,12,45)-(103,215,60)-(776,301,88), the input sample is [103,211,56,889,12, 45,103,215,60], and the output sample is [776,301,88].
[0122] Step S840: Train the second agent using input and output samples to obtain a trained generative model.
[0123] In this embodiment, training is performed by setting a training objective for the second agent, where the training objective is to maximize the log-likelihood. The expression is as follows:
[0124] In the formula, It is the j-th codeword of the next product feature identifier.
[0125] In this embodiment, the accuracy of the second agent is improved by constructing input samples and output samples to train the second agent.
[0126] <Device Embodiment> Figure 7 A schematic diagram of the composition structure of a product display device according to an embodiment of the present disclosure is shown. Figure 7 As shown, the product display device 700 includes an acquisition module 710, a construction module 720, and a determination module 730.
[0127] The acquisition module 710 is used to acquire the buyer's historical browsing information; wherein, the historical browsing information includes the historical products viewed by the buyer, the product attributes representing the historical products and the attribute information of the subordinate assets configured for the historical products, as well as the identifiers of the buyer's interaction behavior with the historical products. The building module 720 is used to invoke the first intelligent agent to construct product preference identifiers for buyer users based on attribute information and interaction behavior identifiers; The determining module 730 is used to invoke the second intelligent agent to determine candidate products that match the product preference identifier in a preset product library; wherein, the candidate products are used to be displayed preferentially when the buyer user browses.
[0128] In some embodiments, the construction module 720 is further configured to invoke a first intelligent agent to construct shared features of historical products and product-side features of shared features under target interactive behavior based on attribute information; determine user-side features based on interactive behavior identifiers; concatenate product-side features and user-side features to obtain fused features; perform semantic transformation on the fused features to construct a first semantic feature representing the buyer user's interactive behavior, and use it as a product preference identifier for the buyer user.
[0129] In some embodiments, the product display device 700 further includes a feature acquisition module, which is used to invoke a first intelligent agent to acquire the user characteristics of the buyer user.
[0130] The building block 720 is also used to determine user-side characteristics based on user features and interaction behavior identifiers.
[0131] In some embodiments, the product display device 700 further includes a first training module, which is used to acquire first user behavior data; wherein, the first user behavior data includes a first user identifier of the first user, a first product identifier of the first product, and a target interaction behavior identifier of the first user's first interaction behavior with the first product; based on the target behavior relationship chain reflected by the target interaction behavior identifier, the sample type of the first user behavior data is determined, and training sample pairs are constructed; through the training sample pairs, the first intelligent agent is trained to obtain a trained first intelligent agent.
[0132] In some embodiments, the product display device 700 further includes an information acquisition module, which is used to invoke a third intelligent agent to acquire input information of the seller user on historical products; and to perform feature transformation processing on the input information to obtain multimodal features.
[0133] The building block 720 is also used to concatenate multimodal features and fused features to obtain the first semantic features that characterize the seller's user interaction behavior.
[0134] In some embodiments, the determining module 730 is further configured to invoke a second intelligent agent to generate a product feature identifier that matches the product characteristics expected by the buyer user based on the product preference identifier; and to determine candidate products corresponding to the product feature identifier in a preset product library.
[0135] In some embodiments, the product display device 700 further includes a second training module, which is used to acquire second user behavior data; wherein, the second user behavior data includes a second product in which the second user interacts, a second product identifier of the second product, and the interaction time with the second product; determines the product semantic feature code mapped by the second product identifier; divides multiple product semantic feature codes into input samples and output samples based on the interaction time; and trains the second agent using the input samples and output samples to obtain a trained generative model.
[0136] <Equipment Example> Figure 8 A schematic diagram of the hardware structure of an electronic device according to some other embodiments is shown. For example... Figure 8 As shown, the electronic device 800 includes a processor 810 and a memory 820. The memory 820 stores a computer program that controls the processor 810 to operate in order to control the electronic device 800 to perform a product display method according to any embodiment of the present disclosure.
[0137] This disclosure also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements a product display method according to any embodiment of this disclosure.
[0138] This disclosure also provides a computer program product, which includes a computer program or instructions that, when executed by a processor, implement a product display method according to any embodiment of the disclosure.
[0139] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. This disclosure can be a system, method, and / or computer program product. A computer program product may include a computer-readable storage medium loaded with computer-readable program instructions for causing a processor to implement any of the methods in the foregoing embodiments of this disclosure. A computer-readable storage medium may be a tangible device capable of holding and storing instructions used by an instruction execution device. For example, a computer-readable storage medium may include an electrical storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device, or any suitable combination thereof. More specific examples of computer-readable storage media (a non-exhaustive list) include: portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM), static random access memory (SRAM), compact disc-read-only memory (CD-ROM), digital versatile disc (DVD), memory sticks, floppy disks, mechanical encoding devices, such as punch cards or recessed protrusions storing instructions thereon, and any combination thereof. Computer-readable storage media as used herein is not to be construed as transient signals themselves, such as radio waves or other freely propagating electromagnetic waves, electromagnetic waves propagating through waveguides or other transmission media (e.g., light pulses through fiber optic cables), or electrical signals transmitted through wires. The computer-readable program instructions described herein can be downloaded from computer-readable storage media to various computing / processing devices, or downloaded via a network, such as the Internet, local area network, wide area network, and / or wireless network, to an external computer or external storage device. The network may include one or more of copper transmission cables, fiber optic transmission, wireless transmission, routers, firewalls, switches, gateway computers, and edge servers. A network adapter card or network interface in each computing / processing device receives the computer-readable program instructions from the network and forwards them to computer-readable storage media in the respective computing / processing device.The computer program instructions used to perform the operations of this disclosure may be assembly instructions, instruction set architecture (ISA) instructions, machine instructions, machine-dependent instructions, microcode, firmware instructions, state setting data, or source or object programs written in any combination of one or more programming languages, including object-oriented programming languages (such as Smalltalk, C++, etc.) and conventional procedural programming languages (such as the "C" language or similar programming languages). The computer-readable program instructions may be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving a remote computer, the remote computer may be connected to the user's computer via any type of network (e.g., a local area network or a wide area network), or it may be connected to an external computer (e.g., via the Internet using an Internet service provider). In some embodiments, electronic circuitry, such as programmable logic circuitry, field-programmable gate arrays, or programmable logic arrays, may be customized by utilizing state information from the computer-readable program instructions to execute the computer-readable program instructions, thereby implementing various aspects of the embodiments of this disclosure. Various aspects of this disclosure are described herein with reference to flowchart illustrations and / or block diagrams of methods, apparatus, and computer program products according to embodiments of this disclosure. It should be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer-readable program instructions. These computer-readable program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, or other programmable data processing apparatus to produce a machine such that, when executed by the processor of the computer or other programmable data processing apparatus, they create means for implementing the functions / actions specified in one or more blocks of the flowchart illustrations and / or block diagrams. These computer-readable program instructions can also be stored in a computer-readable storage medium that causes a computer, programmable data processing apparatus, and / or other device to operate in a particular manner; thus, the computer-readable medium storing the instructions comprises an article of manufacture that includes instructions for implementing aspects of the functions / actions specified in one or more blocks of the flowchart illustrations and / or block diagrams. Computer-readable program instructions may also be loaded onto a computer, other programmable data processing apparatus, or other device to cause a series of operational steps to be performed on the computer, other programmable data processing apparatus, or other device to produce a computer-implemented process, thereby causing the instructions that execute on the computer, other programmable data processing apparatus, or other device to perform the functions / actions specified in one or more boxes of a flowchart and / or block diagram.The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present disclosure. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of an instruction containing one or more executable instructions for implementing a specified logical function. In some alternative implementations, the functions marked in the blocks may occur in a different order than those marked in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or action, or using a combination of dedicated hardware and computer instructions. It should be noted that implementation in hardware, implementation in software, and implementation using a combination of software and hardware are all equivalent. Various embodiments of the present disclosure have been described above; the above description is exemplary and not exhaustive, and is not limited to the disclosed embodiments. Many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of the described embodiments. The terminology used herein is chosen to best explain the principles, practical application, or improvement of the technology in the market, or to enable others skilled in the art to understand the embodiments disclosed herein. The scope of this disclosure is defined by the appended claims.
Claims
1. A method for displaying goods, characterized in that, The method includes: Obtain the buyer's browsing history information; wherein, the browsing history information includes the buyer's browsing history products, attribute information representing the product attributes of the historical products and the subordinate attributes of the subordinate assets configured for the historical products, and the buyer's interaction behavior identifiers with the historical products; The first intelligent agent is invoked to construct a product preference identifier for the buyer user based on the attribute information and the interaction behavior identifier; The second intelligent agent is invoked to determine candidate products that match the product preference identifier from a pre-set product library; wherein, the candidate products are used to be displayed preferentially when the buyer user browses.
2. The method according to claim 1, characterized in that, The invocation of the first intelligent agent, based on the attribute information and the interaction behavior identifier, constructs a product preference identifier for the buyer user, including: The first intelligent agent is invoked to construct the shared features of the historical products and the product-side features of the shared features under the target interaction behavior based on the attribute information. The user-side features are determined based on the interaction behavior identifiers; The product-side features and the user-side features are concatenated to obtain fused features; The fused features are semantically transformed to construct a first semantic feature that represents the buyer user's interaction behavior, and this feature serves as an identifier of the buyer user's product preferences.
3. The method according to claim 1, characterized in that, The method further includes: Invoke the first intelligent agent to obtain the user characteristics of the buyer user; Determining the user-side features based on the interaction behavior identifier includes: The user-side features are determined based on the user characteristics and the interaction behavior identifiers.
4. The method according to claim 1, characterized in that, The method further includes: Acquire first user behavior data; wherein, the first user behavior data includes the first user identifier of the first user, the first product identifier of the first product, and the target interaction behavior identifier of the first user's first interaction behavior with the first product; Based on the target behavior relationship chain reflected by the target interaction behavior identifier, determine the sample type of the first user behavior data and construct training sample pairs; The first agent is trained using the training sample pairs to obtain a trained first agent.
5. The method according to claim 2, characterized in that, The method further includes: Invoke a third-party intelligent agent to obtain the seller user's input information regarding the historical products; The input information is subjected to feature transformation processing to obtain multimodal features; The step of semantically transforming the fused features to construct the first semantic feature includes: The multimodal features and the fused features are concatenated to obtain the first semantic feature representing the seller's user interaction behavior.
6. The method according to claim 1, characterized in that, The invocation of the second intelligent agent, in order to determine candidate products matching the product preference identifier from a pre-set product library, includes: The second intelligent agent is invoked to generate a product feature identifier that matches the product characteristics expected by the buyer user, based on the product preference identifier. In a pre-set product library, candidate products corresponding to the product feature identifier are identified.
7. The method according to claim 6, characterized in that, The method further includes: Acquire second user behavior data; wherein, the second user behavior data includes the second product in which the second user interacts, the second product identifier of the second product, and the interaction time with the second product; Determine the semantic feature encoding of the product mapped to the second product identifier; Based on the interaction time, the multiple semantic features of the products are divided into input samples and output samples; The second agent is trained using the input samples and the output samples to obtain the trained generative model.
8. A product display device, characterized in that, The device includes: The acquisition module is used to acquire the buyer's historical browsing information; wherein, the historical browsing information includes the historical products browsed by the buyer, the product attributes representing the historical products and the attribute information of the subordinate assets configured for the historical products, and the identifiers of the buyer's interaction behavior with the historical products; The construction module is used to invoke the first intelligent agent to construct a product preference identifier for the buyer user based on the attribute information and the interaction behavior identifier; The determination module is used to invoke the second intelligent agent to determine candidate products that match the product preference identifier in a preset product library; wherein, the candidate products are used to be displayed preferentially when the buyer user browses.
9. An electronic device, characterized in that, The system includes a memory and a processor, the memory being used to store a computer program; the processor being used to execute the computer program to implement the method according to any one of claims 1 to 7.
10. A computer-readable storage medium, characterized in that, A computer program is stored on the computer-readable storage medium, which, when executed by a processor, implements the method according to any one of claims 1 to 7.