Information publication method and system, information category determination method, commodity listing method, and device

Abstract

Embodiments of the present disclosure provide an information publication method and system, an information category determination method, a commodity listing method, and a device. According to the technical solution provided by the embodiments of the present disclosure, multi-modal information of a target object is acquired, and a listing category to which the target object belongs is determined on the basis of the multi-modal information. According to the category determination method, mutual complementariness and mutual cross validation of the multi-modal information are utilized, so that a determination result of the listing category has good robustness and is highly accurate. Then, attribute information of the target object can be automatically mapped to attribute information under the listing category, so that automatic loading of object information is implemented, and manual loading operations are reduced. In conclusion, according to the solution, a proper listing category can be determined in real time for the target object to be listed, and the attribute information of the target object is automatically loaded, so that the degree of automation of information publication is increased, and labor costs of information publication can be greatly reduced. The effect of reduction of labor costs is more remarkable especially when information needs to be published in large quantities.

Description

Information publishing, information category determination and product publishing method, system and equipment

[0001] This disclosure claims priority to the Chinese patent application filed with the China Patent Office on November 2, 2023, with application number 202311453309.1 and application name “Methods, systems and equipment for information release, information category determination and product release”, the entire contents of which are incorporated by reference in this disclosure. Technical Field

[0002] The present disclosure relates to the field of e-commerce technology, and in particular to a method, system, and device for information publishing, information category determination, and product publishing. Background Art

[0003] Merchants sell products on e-commerce platforms, and product listing is a crucial step in achieving online sales. During the product listing process, merchants are required to select the product category. Manual category selection is time-consuming and prone to errors, hindering the efficiency and quality of product listings. Furthermore, incorrectly selecting a product category can negatively impact subsequent product promotion and traffic generation, and can also result in a poor consumer experience.

[0004] Summary of the Invention

[0005] In view of the above problems, the present disclosure provides an information publishing, information category determination and product publishing method, product batch publishing method, system and device that solve or at least partially solve the above problems.

[0006] The first embodiment of the present disclosure provides an information publishing method, which includes: obtaining at least one reference object related to the target object based on the multimodal information of the target object; determining at least one target category related to the target object based on the multimodal information of the target object; determining the publishing category to which the target object belongs based on the category to which the at least one reference object belongs and the at least one target category; obtaining target object attribute information corresponding to the publishing category; and using the publishing category and the target object attribute information as the publishing information of the target object to perform a publishing operation.

[0007] A second embodiment of the present disclosure provides a method for determining an information category, which includes: vectorizing the multimodal information of a target object to obtain at least two vectors representing the target object; obtaining multiple preset vectors, wherein one preset vector represents a category or a reference object; determining a preset vector related to the target object based on the at least two vectors representing the target object and the multiple preset vectors; and determining the category to which the target object belongs based on the category represented by the preset vector related to the target object or the category to which the reference object belongs.

[0008] A third embodiment of the present disclosure provides an information publishing method, which includes: in response to a user's operation on a target object, obtaining multimodal information of the target object; based on the multimodal information, displaying the publishing category to which the target object belongs; wherein the publishing category is determined based on the category to which at least one reference object belongs and / or at least one target category, and the at least one reference object and / or the at least one target category are correlated with the multimodal information; displaying the target object attribute information corresponding to the publishing category for verification, and executing the publishing operation on the target object after the verification is passed.

[0009] The fourth embodiment of the present disclosure provides a product publishing method, which includes: in response to a merchant's operation on a product, obtaining multimodal information of the product; based on the multimodal information, displaying the publishing category to which the product belongs; wherein the publishing category is determined based on the category to which at least one reference product belongs and / or at least one target category, and the at least one reference product and / or the at least one target category are correlated with the multimodal information; displaying the product attribute information corresponding to the publishing category for verification, and after the verification is passed, executing the publishing operation for the product, and publishing the product to the e-commerce platform.

[0010] The fifth embodiment of the present disclosure provides an information publishing method, which includes: in response to a user's operation on a target object, obtaining multimodal information of the target object; determining a publishing category to which the target object belongs based on the multimodal information; wherein the publishing category is determined based on a category to which at least one reference object belongs and / or at least one target category, and the at least one reference object and / or the at least one target category are correlated with the multimodal information; obtaining target object attribute information corresponding to the publishing category; performing a publishing operation on the target object based on the publishing category and the target object attribute information; and displaying a preview page after the target object is published.

[0011] The sixth embodiment of the present disclosure provides a product publishing method, which includes: in response to a merchant's operation on a product, obtaining multimodal information of the product; determining a publishing category to which the product belongs based on the multimodal information; wherein the publishing category is determined based on a category to which at least one reference product belongs and / or at least one target category, and the at least one reference product and / or the at least one target category are correlated with the multimodal information; obtaining product attribute information corresponding to the publishing category; performing a publishing operation on the product based on the publishing category and the product attribute information; and displaying a preview page after the product is published.

[0012] The seventh embodiment of the present disclosure provides a method for batch publishing of goods, the method comprising: in response to an operation of triggering batch publishing of goods by a merchant, obtaining multimodal information of multiple goods to be batch published; determining corresponding publishing information for each of the multiple goods; performing a batch publishing operation so that the merchant can batch review the preview pages of the multiple goods after publication; wherein, determining corresponding publishing information for one of the multiple goods comprises: determining a publishing category to which the goods belong based on the multimodal information of the goods; wherein the publishing category is determined based on a category to which at least one reference object belongs and / or at least one target category, and the at least one reference object and / or the at least one target category are correlated with the multimodal information of the goods; and according to the multimodal information of the goods, mounting corresponding product attribute information for the publishing category to which the goods belong to obtain the publishing information of the goods.

[0013] The eighth embodiment of the present disclosure provides an information publishing system. The system includes:

[0014] The client is configured to send multimodal information of a target object to the server in response to a user's operation on the target object;

[0015] The server is configured to obtain, based on the multimodal information, at least one reference object and / or at least one target category that is relevant to the target object; determine, based on the category to which the at least one reference object belongs and / or the at least one target category, a publishing category to which the target object belongs; and determine target object attribute information corresponding to the publishing category;

[0016] The client is used to display the publishing category and the target object attribute information corresponding to the publishing category for verification; in response to the verification pass instruction, it triggers the server to perform a publishing operation on the target object based on the publishing category and the target object attribute information corresponding to the publishing category.

[0017] The ninth embodiment of the present disclosure provides an information publishing system. The system includes:

[0018] The client is configured to send multimodal information of a target object to the server in response to a user's operation on the target object;

[0019] The server is configured to obtain, based on the multimodal information of the target object, at least one reference object and / or at least one target category that is relevant to the target object; determine, based on the category to which the at least one reference object belongs and / or the at least one target category, the publishing category to which the target object belongs; obtain target object attribute information corresponding to the publishing category; and perform a publishing operation on the target object based on the publishing category and the target object attribute information;

[0020] The client is used to display a preview page of the target object after it is published.

[0021] The tenth embodiment of the present disclosure provides an electronic device, which includes a memory and a processor, wherein the memory is used to store a computer program; the processor is coupled to the memory and is used to execute the computer program stored in the memory to implement the steps in the method provided in any of the above method embodiments.

[0022] An eleventh embodiment of the present disclosure provides a computer-readable storage medium storing computer instructions. When the computer instructions are executed on an electronic device, the electronic device executes the method steps recorded in any of the above method embodiments.

[0023] A twelfth embodiment of the present disclosure provides a chip, comprising a processor, wherein the processor is configured to call a computer program in a memory to execute the method steps recorded in any of the above method embodiments.

[0024] A thirteenth embodiment of the present disclosure provides a computer program product, including a computer program, which implements the method steps recorded in any of the above method embodiments when executed by a processor.

[0025] The technical solution provided by the embodiment of the present disclosure obtains the multimodal information of the target object, and determines the publishing category to which the target object belongs based on the multimodal information. This category determination method utilizes the mutual complementarity and mutual cross-validation of multimodal information, so that the publishing category determination result is more robust and more accurate. Afterwards, the target object attribute information can also be automatically mapped to the attribute information under the publishing category, realizing the automatic mounting of the object information and reducing the manual mounting operation. In summary, this solution can determine the appropriate publishing category for the target object that needs to be published in real time, and automatically mount the attribute information of the target object, thereby improving the degree of automation of information publishing and greatly reducing the labor cost of information publishing. Especially when information needs to be published in large quantities, the labor cost reduction effect is more significant.

[0026] In addition, the technical solution provided by the embodiments of the present disclosure also provides a solution for batch publishing of products. Users can import or input multimodal information for multiple products to be published in batches at once. The execution entity will automatically determine the corresponding publishing information for each product and execute the batch publishing operation, allowing merchants to view the preview pages of multiple products after they are published. Users do not need to input or import products one by one to publish them one by one. Using the solution of this embodiment, product publishing efficiency is high. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] In order to more clearly illustrate the embodiments of the present disclosure or the technical solutions in the prior art, a brief introduction will be given below to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present disclosure. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0028] FIG1 is a schematic diagram of the structure of an information publishing system provided by an embodiment of the present disclosure;

[0029] FIG2 is a flow chart of an information publishing method according to an embodiment of the present disclosure;

[0030] FIG3 is a schematic diagram showing the principle of commodity information input according to an embodiment of the present disclosure;

[0031] FIG4 is a schematic diagram showing the principle of determining the category to which a target object belongs according to an embodiment of the present disclosure;

[0032] FIG5 is a flow chart of a method for determining an information category according to an embodiment of the present disclosure;

[0033] 6 and 7 are flowcharts of information publishing methods provided by two other embodiments of the present disclosure;

[0034] 8 and 9 are schematic structural diagrams of an information publishing device provided in an embodiment of the present disclosure;

[0035] FIG10 is a schematic structural diagram of an electronic device provided in an embodiment of the present disclosure. DETAILED DESCRIPTION

[0036] E-commerce merchants often need to list and sell large quantities of products on their respective e-commerce platforms. During this product listing process, existing e-commerce platforms often require merchants to select product categories, which is time-consuming and prone to errors. These issues are exacerbated when releasing large quantities of products. If a product's category is incorrectly selected, it will not only affect subsequent search results and reduce sales, but may also result in incorrect product recommendations to consumers.

[0037] Although some e-commerce platforms have achieved the goal of automatically determining the category to which a product belongs to a certain extent based on pre-trained classification models. However, when building the classification model, only the text information of the product is considered, while other modal information of the product, such as audio, images, videos, etc., is ignored. The text information of the product is often relatively brief. For example, 63% of the product titles have fewer than 10 words, which leads to low category prediction accuracy. In addition, the classification model needs to rely on a large amount of high-quality manually annotated data to conduct end-to-end training in related fields. As a result, the obtained classification model has insufficient generalization performance, lacks autonomous reasoning capabilities, and cannot recognize unseen categories. Once a new category is added to the category structure, it is necessary to manually annotate new sufficient annotated data to retrain the model.

[0038] To address the above issues, the present disclosure provides a technical solution for determining the category of a target object based on its multiple modal information (such as text, image, audio, and video). The target object in the solution provided by the present disclosure can be a commodity. The implementation of the solution in the present disclosure can improve the accuracy of commodity category prediction, further facilitating automatic commodity publishing and thus reducing commodity operating costs for merchants.

[0039] In order to enable those skilled in the art to better understand the solutions of the present disclosure, the technical solutions in the embodiments of the present disclosure will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present disclosure.

[0040] Some processes described in the specification, claims, and figures of this disclosure include multiple operations that appear in a specific order. These operations may be executed in a different order than the order in which they appear in this document, or in parallel. Operation serial numbers, such as 101 and 102, are used solely to distinguish between different operations and do not represent any order of execution. Furthermore, these processes may include more or fewer operations, and these operations may be executed sequentially or in parallel. It should be noted that terms such as "first" and "second" are used herein to distinguish between different messages, devices, modules, etc., and do not represent a sequential order, nor do they limit "first" and "second" to different types. The term "or / and" in this disclosure is merely a description of an association relationship between associated objects, indicating that three possible relationships can exist. For example, "A or / and B" indicates that A can exist alone, A and B can exist simultaneously, or B can exist alone. The character " / " in this disclosure generally indicates that the associated objects are in an "or" relationship. It should also be noted that the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a product or system comprising a series of elements includes not only those elements, but also other elements that are not explicitly listed, or also includes elements that are inherent to such product or system. In the absence of further restrictions, an element defined by the statement "comprising a..." does not exclude the existence of other identical elements in the product or system comprising the element. In addition, the following embodiments are only part of the embodiments of the present disclosure, not all of the embodiments. Based on the embodiments in the present disclosure, all other embodiments obtained by those skilled in the art without making creative work are within the scope of protection of the present disclosure.

[0041] Before introducing the various embodiments of the present disclosure, some terms appearing in this document are briefly introduced.

[0042] Modality is a form of representation of an object. Multimodality typically encompasses two or more modalities, describing an object from multiple perspectives. For example, multimodal information about a product may include, but is not limited to, text, images, audio, and video. Using multimodal data can make product presentations more comprehensive and holistic.

[0043] BLIP-2: Bootstrapping Language-Image Pre-training with Frozen Image Encoders and Large Language Models, a large multimodal model with an extremely large parameter size. BLIP-2 is a versatile and effective pre-training strategy. The various embodiments of this disclosure leverage BLIP-2's image-text multimodal alignment capabilities to infer vectorized representations of information from various modalities. For example, BLIP-2 can be used to infer vectorized representations of product images and textual information, and so on.

[0044] BE: Basic Engine, a recall engine. BE vector recall vectorizes elements (entities) to construct an index for efficient retrieval. On the search side, the search elements (entities) are vectorized in the same way, and retrieval technology is used to retrieve and recall items, obtaining a collection of similar products or elements (entities) and sorting them by distance.

[0045] CPV data, e-commerce platforms use the category-attribute-attribute value (CPV) method to characterize products.

[0046] The various method embodiments provided by the present disclosure can be implemented on the hardware corresponding to the following information publishing system. Specifically, referring to FIG1 , the information processing system includes: a client 100 and a server 200 .

[0047] One specific implementation method is that the client 100 is used to send multimodal information of the target object to the server in response to the user's operation on the target object. The server 200 is used to obtain at least one reference object and / or at least one target category that is relevant to the target object based on the multimodal information; determine the publishing category to which the target object belongs based on the category to which at least one reference object belongs and / or at least one target category; and determine the target object attribute information corresponding to the publishing category. The client 100 is used to display the publishing category and the target object attribute information corresponding to the publishing category for verification; in response to the verification pass instruction, the server is triggered to perform the publishing operation on the target object based on the publishing category and the target object attribute information corresponding to the publishing category.

[0048] Another possible implementation method is for client 100 to send multimodal information of the target object to the server in response to a user's operation on the target object. Server 200 is configured to obtain, based on the multimodal information of the target object, at least one reference object and / or at least one target category that is relevant to the target object; determine the publication category to which the target object belongs based on the category to which the at least one reference object belongs and / or the at least one target category; obtain target object attribute information corresponding to the publication category; and execute the publication operation on the target object based on the publication category and target object attribute information. Client 100 is configured to display a preview page after the target object is published.

[0049] The aforementioned information publishing system can be an e-commerce system. Accordingly, the target objects can be products to be published on the e-commerce platform. Specifically, these can include, but are not limited to: clothing, home furnishings, shoes, bags (such as handbags, backpacks, and luggage), hats, gloves, accessories (such as bracelets, necklaces, bracelets, earrings, and headwear), electronic products (such as watches, mobile phones, laptops, and tablets), bedding, and so on.

[0050] The multimodal information of the target object can be uploaded or input by the user through the client interface.

[0051] The client can be a smartphone, tablet computer, desktop computer, smart wearable device, etc. The server can be a single server, a service cluster, a virtual server deployed on a server, or a cloud.

[0052] The specific contents of the corresponding functions of the hardware such as the client and the server in the information processing system will be described in detail in the following method embodiments.

[0053] FIG2 shows a flow chart of an information processing method provided by an embodiment of the present disclosure. The execution subject of all steps in the method of this embodiment may be the server in the above-mentioned system. Of course, if the hardware device corresponding to the client has strong computing power, the execution subject of the method of this embodiment may also be the client, or the client and the server may execute the method in collaboration (i.e., some steps are executed by the client, and other steps are executed by the server). As shown in FIG2 , the information publishing method provided by the embodiment of the present disclosure includes the following steps:

[0054] 101. Acquire at least one reference object related to the target object based on multimodal information of the target object;

[0055] 102. Determine at least one target category related to the target object based on the multimodal information of the target object;

[0056] 103. Determine a publishing category to which the target object belongs based on the category to which at least one reference object belongs and at least one target category;

[0057] 104. Obtain the target object attribute information corresponding to the release category;

[0058] 105. Use the publishing category and target object attribute information as the publishing information of the target object and perform the publishing operation.

[0059] In this embodiment, the target object mentioned in each step may be the target product (or referred to as a product to be released) that the merchant needs to release through the e-commerce platform. For the type of target product, please refer to the relevant content described in other embodiments above. Accordingly, the multimodal information of the target object is the multimodal information of the product (such as images, audio, video, text, etc.), which can be provided by the merchant or obtained from the network side (such as other platforms). For example, the same e-commerce company has overseas platforms for overseas markets and domestic platforms for the domestic market. If the merchant has released the product on the domestic platform and wants to release the product on the overseas platform as well, it can be obtained from the domestic platform by the executive body of the embodiment of this disclosure.

[0060] In addition to commodities, the target object can also be information (such as news information, courseware information, etc.).

[0061] Various modal information includes, but is not limited to, text information, image information, video information, and audio information. Video information can be converted into image information in an image format, and audio information can be converted into text information in a text format. Therefore, the information included in product information can be roughly divided into the following two modal information types: text information (including text information provided by the merchant and text information obtained by converting audio information into text) and image information (including image information provided by the merchant and image information obtained by extracting video frames from video information).

[0062] Taking the target object as a commodity as an example, the above text information may include the commodity title, commodity attributes (such as material, color, size, etc.), detailed description of the commodity (such as name, brand, specifications, purpose, applicable population, freight, place of origin, etc.), the external website category path corresponding to the commodity, etc. The above image information may include: display pictures of the commodity (such as white background main pictures, scene pictures, detail pictures, etc.), SKU (Stock Keeping Unit inventory in and out measurement unit) pictures (pictures that display the specifications and attributes of the commodity). In specific implementation, merchants can input or upload commodity information through the commodity release page provided by the e-commerce platform. Depending on the input method, merchants can input data of different modes in the commodity information for one or more commodities to be released at one time.

[0063] For example, as shown in FIG3 , a merchant can successfully log in to the merchant backend page provided by an e-commerce platform through the client 100, and enter the product release page shown in FIG3 by clicking the “Publish Product” function control on the merchant backend page. Afterwards, the merchant enters the product information (such as image information, text information) of the target product to be released through the product release page. After completion, the merchant can click the “Publish” control, and the client 100 sends a product release request to the server in response to the merchant’s click operation. After receiving this product release request, the server can obtain the multimodal information of the product provided by the merchant from the information carried in this product release request.

[0064] For another example, referring to Figure 3, it is assumed that the merchant has a demand for batch release of goods, and the e-commerce platform provides a product Excel template for batch release of goods. The merchant downloads the product Excel template on the e-commerce platform where he is stationed, and fills in the product information of multiple goods that need to be released in advance into the product Excel template. Then, the filled-in product Excel template can be imported into the e-commerce platform through the "Upload File" control a in the batch release page provided by the e-commerce platform. In response to this import operation triggered by the merchant, the server can obtain the product Excel template imported by the merchant and obtain the product information (multimodal information) of each of the multiple goods therefrom. In this case, the target object targeted by the above steps of this embodiment can be one of the multiple goods. Multiple goods can be released using the method steps provided in the embodiment of the present disclosure.

[0065] In the above 101, each modal information included in the multimodal information may be vectorized so as to obtain at least one reference object using a vector corresponding to each modal information.

[0066] Similarly, in the above 102, each modal information included in the multimodal information may also be vectorized so as to determine at least one target category using the vector corresponding to each modal information.

[0067] The multimodal information includes at least text information and image information of the target object. Using the computational model, vectorized representations are inferred for the text information and image information of the target object, respectively, to obtain vectors corresponding to the text information and the image information of the target object.

[0068] The above-mentioned computational model can be, but is not limited to, a model obtained by offline training BLIP-2 (Bootstrapping Language-Image Pre-training with Frozen Image Encoders and Large Language Models, a large multimodal model with an extremely large parameter scale) using highly confident manually annotated data. More specifically, during implementation, it can be implemented with the aid of RTP-PY (Real-Time Processing in Python) based on BLIP-2. RTP-PY is a real-time reasoning service platform based on the Python language. The above-mentioned manually annotated data is the data of the annotated reference object (also known as the anchor object, such as the anchor product), and the category to which the reference object belongs has high accuracy. When training BLIP-2 offline, the training can be implemented specifically using the text information and image information of the reference object.

[0069] The text and image information of the target object can be preprocessed first, and then the deployed computing model is called to perform online inference on the preprocessed text and image information of the target object to obtain the corresponding vectorized representation. Specifically:

[0070] Preprocessing the target text information may include: if the text information is in a non-default language, translating the text information into the default language. The translated text information is then normalized, including but not limited to case conversion, accent handling, special punctuation, singular / plural handling, and error correction. This preprocessed text information is then input into a vectorized representation processing model, which then uses the vectorized representation processing model to online infer the corresponding text vector.

[0071] Preprocessing of the image information of the target object may include: if the image information is represented by a corresponding link address URL (Universal Resource Locator), then the link address URL corresponding to the image information may be converted into a corresponding transmission format such as BASE64 (BASE64 Encoding) for input to RTP-PY, so that RTP-PY can obtain the corresponding image information based on the received link address, thereby online inferring the image vector corresponding to the image information.

[0072] That is, further, in this embodiment, step 101 of "obtaining at least one reference object related to the target object based on the multimodal information of the target object" can be implemented by the following steps:

[0073] 1011. Vectorize the multimodal information of the target object to obtain at least two vectors representing the target object;

[0074] 1012. Obtain object information of a plurality of preset reference objects, wherein the object information of a reference object includes at least two vectors representing the reference object;

[0075] 1013. Determine the correlation between the target object and the reference object based on at least two vectors representing the target object and at least two vectors representing the reference object;

[0076] 1014. Obtain at least one reference object whose relevance meets a first requirement from a plurality of preset reference objects.

[0077] In the above-mentioned 1011, the multimodal information of the target object may include, but is not limited to, text information and image information. Of course, multimodal information may also include audio information, video information, and so on. The audio information, video information, and so on can be processed into text information and image information through a pre-processing process. Then, a computational model is used to generate a vectorized representation corresponding to the text information, namely, a first text vector; and a vectorized representation corresponding to the image information, namely, a first image vector, is generated using the computational model. The computational model may be a BLIP-2 model, or, in specific implementations, may utilize RTP-PY, which implements the vectorized reasoning representation of the multimodal information.

[0078] In the above step 1012, the object information of the preset multiple parameter objects can be pre-derived using a computational model, more specifically, using the zero-shot learning capability of BLIP-2, to perform offline inference on the text and image information of the reference object to generate a corresponding vectorized representation. Based on the vectorized representation, an index data table (such as the reference object data table shown in FIG. 4 ) is constructed for the reference object. Table 1 below shows an example of a reference object data table.

[0079] Table 1 Reference object data table

[0080] After obtaining the reference object data table, a vector similarity calculation method, such as cosine similarity, Manhattan distance, etc., can be used to calculate the similarity between the first text vector corresponding to the target object and the second text vector corresponding to each reference object in the reference object data table, and / or calculate the similarity between the first image vector of the target object and the second image vector corresponding to each reference object in the reference object data table, thereby determining the similarity between the target object and each reference object based on the similarity results calculated above, and based on the similarity between the target object and each reference object, selecting at least one reference object closest to the target object from multiple reference objects. For example, the top N reference objects closest to the target object are selected from multiple reference objects, where N can be a positive integer greater than 1. That is, the above 1013 can be performed using a corresponding recall engine (such as shown in Figure 4) during specific implementation. The recall engine can be, but is not limited to, a BE (Behavior Engine, a recommendation recall engine implemented based on a DII (Data Integration and Inference) algorithm online service platform). Specifically, the recall engine can be used to perform vector similarity calculation on the first text vector and the first image vector corresponding to the target object and the second text vector and the second image vector corresponding to each reference object in the index data table pre-constructed for the reference object (i.e., the reference object data table in Figure 4), thereby selecting a reference object similar to the target object based on the vector similarity calculation results.

[0081] That is, the above-mentioned 1013 “determining the correlation between the target object and the reference object based on at least two vectors representing the target object and at least two vectors representing the reference object” may include the following steps:

[0082] S11, determining a first similarity between a target object and a reference object based on the first text vector and the second text vector;

[0083] S12, determining a second similarity between the target object and the reference object based on the first image vector and the second image vector;

[0084] S13: Determine the correlation between the target object and the reference object based on the first similarity and the second similarity.

[0085] Assume that the first text vector corresponding to the target object is A w , the second text vector corresponding to the i-th reference object is B w i, then A w With B w The similarity between i can be expressed using the cosine similarity formula shown below:

[0086] The first image vector corresponding to the target object is A t , the second image vector corresponding to the i-th reference object is B ti , then A t With B ti The similarity between them can be expressed using the cosine similarity formula shown below:

[0087] The first similarity between the first text vector and the second text vector corresponding to the i-th reference object is recorded as cosθ wwi The second similarity between the first image vector and the second image vector corresponding to the i-th reference object is cosθ tti .

[0088] Accordingly, the above S13 “determining the relevance between the target object and the reference object according to the first similarity and the second similarity” may specifically be:

[0089] The weighted similarity of the first similarity and the second similarity is calculated through weighted calculation, and the weighted similarity result is used as the correlation between the target object and the reference object.

[0090] That is, correlation = Mcosθ wwi +Ncosθ tti , where M and N are weighted values, and the specific values ​​are not limited in this embodiment and can be set manually according to actual conditions or calculated based on a preset algorithm.

[0091] Furthermore, the method provided in this embodiment may further include the following steps:

[0092] S14, determining a third similarity between the target object and the reference object based on the first text vector and the second image vector;

[0093] S15. Determine a fourth similarity between the target object and the reference object based on the first image vector and the second text vector;

[0094] Correspondingly, when determining the correlation between the target object and the reference object in S13 above, the correlation is determined based on the first similarity, the second similarity, the third similarity, and the fourth similarity.

[0095] Assume that the first text vector corresponding to the target object is A w , the second image vector corresponding to the i-th reference object is B ti , then A w With B ti The similarity between them can be expressed using the cosine similarity formula shown below:

[0096] The first image vector corresponding to the target object is A t, the second text vector corresponding to the i-th reference object is B w i, then A t With B w The similarity between i can be expressed using the cosine similarity formula shown below:

[0097] Correspondingly, the correlation between the target object and the reference object == Mcosθ wwi +Ncosθ tti +Ccosθ wti +Dcosθ twi , where M, N, C and D are weighted values, and the specific values ​​are not limited in this embodiment and can be set manually according to actual conditions or calculated based on a preset algorithm.

[0098] According to the above steps, the correlation between the target object and each reference object can be calculated. Afterwards, based on the correlation between the target object and each reference object, the reference objects can be sorted by size according to the correlation value, so as to select the reference objects ranked in the top N. These top N reference objects are the reference objects related to the target object. Alternatively, reference objects whose correlation values ​​are greater than a preset similarity threshold can be selected and determined as reference objects related to the target object. The above N can be flexibly set according to actual conditions, and this embodiment does not make specific restrictions on this. That is, the "first requirement" in step 1014 above can be but is not limited to: "reference objects ranked in the top N", or "the correlation value is greater than the preset similarity threshold", etc.

[0099] After obtaining at least one reference object related to the target object, since the reference object is an object manually marked with high confidence, the accuracy of the category to which the marked reference object belongs can be ensured. Therefore, when determining the category to which the target object belongs, the category to which at least one reference object related to the target object belongs can be referred to.

[0100] The category to which the reference object belongs can be a leaf category. A leaf category is the last level in the pre-set category structure (category tree) and cannot be further subdivided. For example, in an e-commerce platform scenario, categories are generally divided into four levels: primary, secondary, tertiary, and leaf categories (fourth level). For example, the primary level is agricultural and sideline products, the secondary level is aquatic products, the tertiary level is fish, and the leaf category is freshwater fish.

[0101] In an achievable example, step 102 of “determining at least one target category related to the target object based on the multimodal information of the target object” in this embodiment may include the following steps:

[0102] 1021. Vectorize the multimodal information of the target object to obtain at least two vectors representing the target object;

[0103] 1022. Obtain category information of a plurality of preset categories, wherein the category information includes vectors representing the preset categories;

[0104] 1023. Determine the correlation between the target object and the preset category based on at least two vectors representing the target object and the vector representing the preset category;

[0105] 1024. Obtain at least one target category from a plurality of preset categories, the relevance of which meets the second requirement.

[0106] In the above-mentioned 1021, the multimodal information of the target object may include, but is not limited to, text information and image information. Of course, multimodal information may also include audio information, video information, and so on. The audio information, video information, and so on can be processed into text information and image information through a pre-processing process. Then, a computational model is used to generate a vectorized representation corresponding to the text information, namely, a first text vector; and a computational model is used to generate a vectorized representation corresponding to the image information, namely, a first image vector. The computational model may be a BLIP-2 model, or, in specific implementations, may utilize RTP-PY to implement vectorized reasoning representation of the multimodal information.

[0107] That is, 1021 may specifically include: inputting the multimodal information of the target object into the computing model, and executing the computing model to output at least two vectors representing the target object.

[0108] In the specific implementation of the above 1022, the category information of multiple preset categories can be obtained in advance using a calculation model. That is, the text information of each preset category is input into the calculation model, and the calculation model is executed to output a vector representing each preset category. More specifically, the zero-shot capability of BLIP-2 is used to perform offline inference on the text information of each category in the preset category structure to obtain a corresponding vectorized representation, and then an index data table (such as the category data table in Figure 4) is constructed for each category in the category structure based on the vectorized representation result. In the process of offline inference on the text information of the category to obtain a corresponding vectorized representation, the text information can also be expanded to enrich the text description content of the category. For example, the text information can be expanded based on a natural language processing model driven by artificial intelligence technology, and at the same time, typical category words under the category can be added to the text information. The above natural language processing model can be, for example, ChatGPT (Chat Generative Pre-trained Transformer, chatbot).

[0109] That is, the embodiment of the present disclosure may further include the following steps:

[0110] 106. Get text information of preset categories;

[0111] 107. Generate consulting questions based on text information;

[0112] 108. Use consultation questions to trigger the natural language processing model to expand the text information of preset categories;

[0113] 109. Perform vectorization on the expanded text information of the preset category to obtain a vector representing the preset category.

[0114] Table 2 below shows an example of a category data table.

[0115] Table 2 Category data table

[0116] If the category structure changes, this embodiment only needs to use the computing model to infer the new category, perform vectorized representation, and store it, without making any other modifications.

[0117] After having the category data table, you can use vector similarity calculation methods, such as cosine similarity, Manhattan distance, etc., to calculate the similarity between the first text vector corresponding to the target object and the vector representing the preset category (such as the third text vector). For the specific similarity calculation implementation, please refer to the relevant content described above for the reference object, which will not be described in detail here. Afterwards, sort by similarity from large to small, and select the top N preset categories, that is, at least one target category related to the target object. Alternatively, select a preset category with a similarity value greater than a set threshold as the target category related to the target object.

[0118] In addition to calculating the similarity between the first text vector of the target object and the vector representing the preset category, the similarity between the first image vector of the target object and the vector representing the preset category may also be calculated. That is, the aforementioned step 1023 "determining the relevance between the target object and the preset category based on the at least two vectors representing the target object and the vector representing the preset category" may include:

[0119] S21. Determine a first similarity between the target object and the category based on the first text vector and the vector representing the preset category;

[0120] S22. Determine a second similarity between the target object and the category based on the first image vector and the vector representing the preset category;

[0121] S23: Determine the relevance between the target object and the preset category based on the first similarity and / or the second similarity.

[0122] That is, the first method is to directly use the first similarity as the correlation between the target object and the preset category. The second method is to directly use the second similarity as the correlation between the target object and the preset category. The third method is to calculate the weighted average of the first similarity and the second similarity and use the weighted average as the correlation between the target object and the preset category.

[0123] Assume that the first text vector corresponding to the target object is A w , the third text vector corresponding to the j-th preset category is C wj , then A w with C wj The similarity between them can be expressed using the cosine similarity formula shown below:

[0124] The first image vector corresponding to the target object is A t , the third text vector corresponding to the j-th preset category is C wj , then A t with C wj The similarity between them can be expressed using the cosine similarity formula shown below:

[0125] If the third method is used, the correlation between the target object and the jth preset category = aMcosθ wwj +bcosθ twj , where a and b are weighted values ​​respectively. The specific values ​​are not limited in this embodiment and can be set manually according to actual conditions or calculated based on a preset algorithm.

[0126] The above process can be completed using a corresponding recall engine (such as the example shown in FIG4 ). The recall engine can be, but is not limited to, BE (a recommendation recall engine implemented based on the DII algorithm online service platform).

[0127] Furthermore, in this embodiment, step 103 of "determining the publishing category to which the target object belongs based on the category to which at least one reference object belongs and at least one target category" can be implemented by the following steps:

[0128] 1031. Merge the category to which at least one reference object belongs and at least one target category to obtain a candidate category list without duplicate categories;

[0129] 1032. Sort the categories in the candidate category table;

[0130] 1033. The category arranged in the set position is used as the publishing category to which the target object belongs.

[0131] For ease of explanation, the category to which at least one reference object belongs is referred to as a first candidate category set, and the at least one target category is referred to as a second candidate category set. In specific implementation, when performing category deduplication and fusion processing, duplicates will be removed for the same category.

[0132] For example, if the first candidate category set includes {category a (correlation value a), category b (correlation value b), category c (correlation value c), category d (correlation value d)}; and the second candidate category set includes {category a (correlation value a'), category c (correlation value c'), category e (correlation value e'), category f (correlation value f')}, then deduplication is required during the fusion process to obtain a candidate category list without duplicate categories:

[0133] {Category a (correlation value a, correlation value a'), category b (correlation value b), category c (correlation value c, correlation value c'), category d (correlation value d), category e (correlation value e'), category f (correlation value f')}

[0134] The correlation values ​​following each of the above categories are calculated using the methods described above. For details, please refer to the above content and will not be repeated here. Some categories have only one correlation value, while others have two. For categories with two correlation values, one correlation value is calculated using S11-S12 or S11-S14 above, and the other correlation value is calculated using S21-S23 above.

[0135] After obtaining the above-mentioned candidate category table without duplicate categories, each category can be sorted based on the correlation value corresponding to each category in the candidate category table. A category with two correlation values ​​indicates that it exists in both candidate category sets, and this category has a greater correlation with the target object. Therefore, when sorting, some similarity scoring algorithms can be used to sort the categories in the candidate category table. The simplest way is to simply add the correlation values ​​of each category to obtain the final correlation value, and then sort it. To make it more complicated, a similarity scoring algorithm can be designed to add an additional point to categories with two correlation values ​​to improve their sorting order, and so on. This embodiment does not limit this.

[0136] The above 1033 may specifically be: using the first ranked category as the publishing category of the target object.

[0137] Usually, the publication is completed only after the attribute information of the target object is mounted on the corresponding leaf category. Corresponding category attribute information is set under each leaf category. The category attribute information can be obtained through the attribute template mounted on the leaf category, and the target object attribute information is mapped one by one to the attribute value of the corresponding attribute under the target category, that is, the purpose of mounting the target object attribute information on the target category is completed. Therefore, in an implementable solution, the above 104 "obtaining the target object attribute information corresponding to the publishing category" can specifically include:

[0138] 1041. Get the attribute template under the publishing category;

[0139] 1042. Fill the target object attribute information into the appropriate position in the attribute template.

[0140] It should be noted here that after mapping the target object attributes to the attribute information under the publishing category, the attribute values ​​of each attribute under the publishing category can also be verified, such as data standardization verification, publisher (merchant) business object range verification, etc. Once the verification is passed, the target object is released.

[0141] The technical solution provided in this embodiment obtains the multimodal information of the target object, and determines the publishing category to which the target object belongs based on the multimodal information. This category determination method utilizes the mutual complementarity and cross-validation of multimodal information, making the publishing category determination result more robust and more accurate. Afterwards, the target object attribute information can also be automatically mapped to the attribute information under the publishing category, realizing the automatic mounting of the object information and reducing the manual mounting operation. In summary, this solution can determine the most appropriate publishing category for the target object to be published in real time, and automatically mount the attribute information of the target object, thereby improving the degree of automation of information publishing and greatly reducing the labor cost of information publishing. Especially when information needs to be published in large quantities, the labor cost reduction effect is more significant.

[0142] Furthermore, the method provided in this embodiment may further include the following steps:

[0143] 110. When a new category is added to the preset category structure, the text information of the new category is obtained;

[0144] 111. Expand the text information of the new category to obtain expanded text information;

[0145] 113. Perform vectorization on the text vector corresponding to the expanded text information to obtain a vector corresponding to the new category;

[0146] 114. Add the new category and the vector corresponding to the new vector to the category data table.

[0147] Through the above steps 110 to 114, the category structure can be changed. For example, when a new category is added, it is only necessary to use the computational model to infer the corresponding vectorized representation of the new category and store it without making any other modifications, such as re-annotating data for the new category to retrain the corresponding model.

[0148] In summary, the technical solution provided by this disclosure leverages BLIP2's multimodal alignment capabilities for images and text to build a complete real-time information publishing system. Taking the product release scenario of an e-commerce platform as an example, the technical solutions provided by this disclosure have built a real-time product release system, providing the technical capabilities for accurate and rapid product release, thereby helping e-commerce platforms rapidly expand their product offerings.

[0149] The specific steps of the entire system process are as follows:

[0150] Step 1: Before launching online smart products, you need to prepare offline data. This stage includes:

[0151] 1. Manually label product data with high confidence to ensure that the published categories of the labeled products are accurate.

[0152] 2. Leveraging BLIP2's zero-shot capabilities, we infer the text descriptions and product images of high-confidence products (manually labeled anchor products), vectorize their representation, and create a corresponding anchor product index data table (same as the reference object data table in Figure 4).

[0153] 3. Using the zero-shot capability of BLIP2, the text description of the category (expanded based on chatGPT and supplemented with typical category words under the category) is inferred offline and vectorized, and the corresponding category index data table is established (the same as the category data table in Figure 4).

[0154] It should be added here that index back to the table means that when executing a query statement, the MySQL database needs to first search the index table, and then search the main table for the corresponding data according to the pointer on the index table.

[0155] Step 2: For the multimodal information of the product to be released, perform corresponding preprocessing according to the information of different modes, including:

[0156] 1. For text-based product information (such as product titles, attributes, detailed descriptions, and external website category paths), if it needs to be translated into the default language, it must be translated first; if it does not need to be translated, this step can be skipped. The translated text is normalized (such as converting uppercase to lowercase, handling accents, special punctuation, singular and plural processing, and error correction).

[0157] 2. For image-based product information (such as product display images and SKU (Stock Keeping Unit) images), convert the image storage URL into BASE64 format. The preprocessing results of different modalities serve as input for the next step of online product vector inference.

[0158] Step 3: Call the deployed BLIP2 multimodal large-scale online service to infer vectorized representations of the product's text description and image, and then perform the recall, sorting, and re-ranking phase:

[0159] Recall ranking: Based on the product's multimodal (text / image) large model vectorized representation, the top N anchor products closest to the target product are recalled from multiple anchor products based on cosine similarity (see above for the specific process). Based on the product's multimodal (text / image) large model vectorized representation, the top N leaf categories closest to the target product are retrieved based on cosine similarity.

[0160] Rearrangement: The categories of the top N anchor products and the top N leaf categories are dynamically merged based on similarity scores to generate a rearranged predicted category ranking. Finally, the top 1 category is obtained as the product release category.

[0161] Step 4: Based on the product release category, map the CPV numbers in the product attribute information input to the attributes and attribute values ​​under the category.

[0162] Step 5: Finally, verify the fields of the product to be released (such as data standardization verification, business scope verification of the seller who releases the product, etc.). Once the verification is completed, the product is released.

[0163] This solution is based on the large model BLIP2 with a large number of parameters, so it has excellent generalization capabilities. Compared with end-to-end modeling solutions, it has two obvious advantages:

[0164] 1. It makes full use of the multimodal information of the product's text and images, complements each other, and cross-validates, making the prediction results more accurate and robust.

[0165] 2. When the category structure changes, only large-scale wiring is required to re-infer and store the new categories, without any other modifications. Traditional end-to-end models require re-labeling the data and retraining the model.

[0166] Another embodiment of the present disclosure provides a method for determining information categories. Figure 5 shows a flow chart of the method for determining information categories provided by an embodiment of the present disclosure. The execution entity of all steps in the method can be the server in the above-mentioned system embodiment. Referring to Figure 5, the information processing method provided by this embodiment includes the following steps:

[0167] 201. Vectorize the multimodal information of the target object to obtain at least two vectors representing the target object;

[0168] 202. Acquire multiple preset vectors, where each preset vector represents a category or a reference object;

[0169] 203. Determine a preset vector related to the target object based on at least two vectors representing the target object and a plurality of preset vectors;

[0170] 204. Determine the category to which the target object belongs based on the category represented by the preset vector related to the target object or the category to which the reference object belongs.

[0171] The preset vectors in the above 202 include: at least two vectors representing the reference object obtained by vectorizing the multimodal information of the reference object; and / or vectors representing the preset category obtained by vectorizing the text description of the preset category.

[0172] In the above 201 and 202 , the vectorized representation process can be implemented by referring to the solution in the above embodiment, such as using the BLIP2 model.

[0173] The specific implementation of the above 203 and 204 can also be found in the corresponding content above, which will not be repeated here.

[0174] Furthermore, the method provided in this embodiment may further include the following steps:

[0175] 205. Obtain target object attribute information corresponding to the category to which the target object belongs;

[0176] 206. Use the category to which the target object belongs and the target object attribute information as the target object's publishing information and perform a publishing operation.

[0177] Similarly, the specific implementation of the above steps 205 and 206 can be found in the corresponding content above, and will not be repeated here.

[0178] It should be noted here that the information category determination method embodiment provided in this embodiment, in addition to including the above steps, may also include other steps mentioned above.

[0179] Furthermore, the present disclosure also provides an embodiment of an information publishing method. As shown in FIG6 , the execution subject of the method provided in this embodiment can be the client in the above system embodiment. Specifically, the method includes:

[0180] 301. In response to a user operation on a target object, obtain multimodal information of the target object;

[0181] 302. Based on the multimodal information, display a publication category to which the target object belongs; wherein the publication category is determined based on a category to which at least one reference object belongs and / or at least one target category, and the at least one reference object and / or at least one target category are correlated with the multimodal information;

[0182] 303. Display the target object attribute information corresponding to the release category for verification. After the verification is passed, execute the release operation for the target object.

[0183] One feasible technical solution is that the above step 302 of “displaying the publishing category to which the target object belongs based on the multimodal information” may include:

[0184] 3021. Obtain object information of a plurality of preset reference objects and category information of a plurality of preset categories;

[0185] 3022. Acquire at least one reference object related to the target object from the multiple reference objects based on the multimodal information and the object information of the multiple reference objects;

[0186] 3023. Acquire at least one target category related to the target object from a plurality of preset categories based on the multimodal information;

[0187] 3024. Determine a publishing category to which the target object belongs based on the category to which at least one reference object belongs and at least one target category.

[0188] A second feasible technical solution is that the publication category to which the target object belongs is determined based on the category to which at least one reference object belongs. For example, the above 302 "displaying the publication category to which the target object belongs based on multimodal information" may include:

[0189] 3021', obtaining object information of a plurality of preset reference objects;

[0190] 3022′, acquiring at least one reference object related to the target object from the multiple reference objects according to the multimodal information and the object information of the multiple reference objects;

[0191] 3023'. Determine the publishing category to which the target object belongs based on the category to which at least one reference object belongs.

[0192] For example, the category to which a reference object most relevant to the target object belongs is determined as the publishing category to which the target object belongs.

[0193] A third feasible technical solution is that the publication category to which the target object belongs is determined based on at least one target category. For example, the above 302 "displaying the publication category to which the target object belongs based on multimodal information" may include:

[0194] 3021", obtain category information of multiple preset categories;

[0195] 3022”, obtaining at least one target category related to the target object from a plurality of preset categories according to the multimodal information;

[0196] 3023”, based on at least one target category, determine the publishing category to which the target object belongs.

[0197] For example, a target category that is most relevant to the target object is determined as the publishing category to which the target object belongs.

[0198] In the above 302, it is taken into account that when the publisher (such as a merchant) inputs the information of the target object, the input information may also include the publishing category to which the target object belongs. In order to protect the rights and interests of the publisher, when the client or server determines that the publishing category of the target object is inconsistent with the input publishing category, the publishing category automatically determined by the client or server for the target object will be displayed on the client interface corresponding to the publisher, allowing the publisher to participate in confirming the selection. The publisher can use voice, keyboard, mouse, etc. to trigger the corresponding confirmation selection operation for the publishing category. For example, you can insist on selecting what you input, or you can select the publishing category automatically confirmed by the system for the target object. For the specific implementation description of mapping the target object attribute information to the attribute information under the publishing category, please refer to the relevant content in other embodiments above.

[0199] Taking information as a commodity as an example, another embodiment of the present disclosure provides a commodity publishing method. Specifically, the commodity publishing method includes:

[0200] 401. Responding to a merchant's operation on a product, obtaining multimodal information of the product;

[0201] 402. Displaying a release category to which the product belongs based on the multimodal information; wherein the release category is determined based on a category to which at least one reference product belongs and / or at least one target category, and the at least one reference product and / or at least one target category is correlated with the multimodal information;

[0202] 403. Display the product attribute information corresponding to the release category for verification. After the verification is passed, execute the release operation for the product and release the product to the e-commerce platform.

[0203] It should be noted that, in addition to the above steps, this embodiment also includes the other steps mentioned above, the specific contents of which can be found in the above content and will not be described in detail here. In addition, the specific implementation contents of the above steps can also be found in the above content.

[0204] In the embodiment of the information publishing method shown in Figure 6 above, the user (information publisher) will first enter the product publishing page as shown in Figure 3, and then enter the product information (i.e., the multimodal information of the target object) through the product publishing page. The publishing category automatically determined by the system for the target object can be displayed on the product publishing page, which the user can see and verify. Click "Publish" after the verification is passed. There is also a case of batch publishing. In this case, the user may only need to upload the object file for batch publishing. The system will automatically publish all the target objects in the file together. The user will not be aware of the entire publishing process, that is, there is no product publishing page like the one on the left in Figure 3 for the user to verify. For such a scenario, the present disclosure provides an embodiment of an information publishing method, as shown in Figure 7, which includes:

[0205] 401. In response to a user operation on a target object, obtain multimodal information of the target object;

[0206] 402. Determine a publication category to which the target object belongs based on the multimodal information; wherein the publication category is determined based on a category to which at least one reference object belongs and / or at least one target category, and the at least one reference object and / or at least one target category are correlated with the multimodal information;

[0207] 403. Obtain the target object attribute information corresponding to the publishing category;

[0208] 404. Execute a publishing operation on the target object according to the publishing category and the target object attribute information;

[0209] 405. Display the preview page of the target object after it is published.

[0210] The specific implementation of the above 401 to 404 can refer to the corresponding content above, which will not be described in detail here. Similarly, in addition to the above steps, this embodiment also includes other steps mentioned above, and the specific contents can refer to the above contents.

[0211] In this embodiment, the user is unaware of the publishing process, and can view the preview page after the final target object is published to check for omissions, correct errors, etc.

[0212] Similarly, taking information as a commodity as an example, the present disclosure also provides multiple commodity publishing method embodiments. Specifically:

[0213] A product publishing method provided by an embodiment of the present disclosure includes:

[0214] 501. In response to a merchant's operation on a product, obtain multimodal information of the product;

[0215] 502. Determine a release category to which the product belongs based on the multimodal information; wherein the release category is determined based on a category to which at least one reference product belongs and / or at least one target category, and the at least one reference product and / or at least one target category are correlated with the multimodal information;

[0216] 503. Obtain product attribute information corresponding to the published category;

[0217] 504. Execute the product publishing operation based on the publishing category and product attribute information;

[0218] 505. Display the preview page after the product is released.

[0219] It should be noted that, in addition to the above steps, this embodiment also includes the other steps mentioned above, the specific contents of which can be found in the above content and will not be described in detail here. In addition, the specific implementation contents of the above steps can also be found in the above content.

[0220] Another embodiment of the present disclosure provides a method for publishing products in batches, including:

[0221] 601. In response to a merchant triggering a batch release of products, obtain multimodal information of multiple products to be released in batch;

[0222] 602. Determine corresponding release information for each of the multiple products;

[0223] 603. Execute a batch publishing operation so that merchants can view preview pages of multiple products after they are published.

[0224] The step 602 of determining corresponding release information for one of the multiple products may include:

[0225] Determining a publication category for the product based on the product's multimodal information; wherein the publication category is determined based on a category to which at least one reference object belongs and / or at least one target category, and the at least one reference object and / or at least one target category are correlated with the product's multimodal information;

[0226] According to the multimodal information of the product, the corresponding product attribute information is mounted for the release category to which the product belongs to obtain the release information of the product.

[0227] It should be noted that the execution entity of this embodiment can be the client or server in the above-mentioned system embodiment. Alternatively, the execution entity of 601 above is the client, and the execution entity of 602 and 603 is the server. In addition, in addition to the above-mentioned steps, this embodiment also includes the other steps mentioned above. The specific details can be found above and will not be repeated here. The specific implementation content of each of the above steps can also be found above.

[0228] Furthermore, the above step 601 of "obtaining multimodal information of multiple products to be released in batches in response to the merchant triggering the operation of batch release of products" may include:

[0229] Display the batch release page;

[0230] In response to an input operation of a merchant on a batch publishing page, obtaining a batch publishing file input by the merchant;

[0231] Read multimodal information of multiple products from a batch publishing file.

[0232] In the above description, the batch publishing page can be similar to the page shown on the right in Figure 3. Merchants can click the "Upload File" control a on the batch publishing page to import or enter the batch publishing file, allowing the client to read the multimodal information of multiple products from the batch publishing file and send it to the server. Alternatively, the client can directly send the batch publishing file to the server. Based on the multimodal information of multiple products, the server can determine the corresponding publishing information for each of them and execute the batch publishing operation. The server then provides the client with a preview page of the published products.

[0233] For example, the client interface displays a preview page after one of the products is released. The client interface can be controlled to display the preview page after the next product is released through touch operations (such as sliding operations) or controls (such as the next control). In order to facilitate merchants to modify or provide feedback, editing and / or feedback controls are also displayed on the preview page. For example, the merchant clicks the edit control, and the preview page displayed on the current interface is editable, and the merchant can modify it. Alternatively, the merchant clicks the feedback control, and a feedback pop-up window is displayed on the current interface, and the merchant can fill in the modification opinions in the feedback pop-up window. The modification opinions can be fed back to the manual platform and adjusted by the manual platform. Alternatively, the modification opinions are fed back to the server, and the server modifies the preview page based on the feedback opinions. Furthermore, a confirmation control can also be displayed on the preview page, and the merchant clicks the confirmation control to confirm. In response to the merchant triggering confirmation for a preview page, the client sends a publishing instruction to the server to instruct it to be published on the e-commerce platform.

[0234] It's important to note that the product preview page mentioned in this article is not the final page published on the e-commerce platform. Only after the merchant confirms the product, the server will send the preview page as the published page to the e-commerce platform's corresponding server.

[0235] In some cases, merchants may tag their products with their own product category information. While merchants may have a relatively clear understanding of the product category, the corresponding category on the e-commerce platform may differ due to differences in category structure or definition.

[0236] Since the product categories determined by the user are relatively accurate, during batch processing, products with the same product category information can be grouped together. When determining the product's release category, only the technical means provided in the above embodiments need to be used to determine the release category of one product. Other products in the same group will also belong to that release category. This significantly reduces the number of product release categories determined by the server, reduces the amount of computation required, and can shorten the batch release response time.

[0237] That is, if the multiple products include products of different categories; the multimodal information of the products contains product category information. In response, the product batch release method provided in this embodiment may also include the following steps:

[0238] Group products with the same product category information into one group;

[0239] When determining the publishing category to which a product belongs based on the multimodal information of the product, the publishing category to which one product in a group belongs is determined, and other products in the same group belong to the same publishing category.

[0240] For example, if product 1, product 4, and product 5 have the same product category information, you only need to determine the release category of product 1 based on the multimodal information of product 1. If product 1 is determined to be in category A, the release categories of products 4 and 5 are also A.

[0241] From the user's (or merchant's) perspective, once the user inputs or imports the multimodal information of multiple products to be released in batches, and clicks the "Batch Release" control, the client and / or server will automatically execute the product release operation; the user can intuitively feel the effect of "one-click batch release", which is simple to operate and highly efficient.

[0242] The technical solution provided by this embodiment allows users to import or enter multimodal information for multiple products to be published in batches at once. The execution entity automatically determines the corresponding publishing information for each product and executes the batch publishing operation, allowing merchants to view preview pages of multiple products after they are published. Users no longer need to enter or import products one by one to publish them one by one. Using this embodiment's solution, product publishing efficiency is high.

[0243] Figure 8 shows a structural diagram of an information publishing device provided by an embodiment of the present disclosure. As shown in Figure 8, the information publishing device includes: an acquisition module 11, a determination module 12 and an execution module 13. Among them, the acquisition module 11 is used to acquire at least one reference object related to the target object based on the multimodal information of the target object. The determination module 12 is used to determine at least one target category related to the target object based on the multimodal information of the target object; and is also used to determine the publishing category to which the target object belongs based on the category to which at least one reference object belongs and at least one target category. The acquisition module is also used to obtain target object attribute information corresponding to the publishing category. The execution module 13 is used to use the publishing category and the target object attribute information as the publishing information of the target object to perform a publishing operation.

[0244] Furthermore, when acquiring at least one reference object related to the target object based on the multimodal information of the target object, the acquisition module 11 is specifically configured to:

[0245] Vectorizing the multimodal information of the target object to obtain at least two vectors representing the target object;

[0246] Acquiring object information of a plurality of preset reference objects, wherein the object information of the reference objects includes at least two vectors representing the reference objects;

[0247] determining a correlation between the target object and the reference object based on at least two vectors representing the target object and at least two vectors representing the reference object;

[0248] At least one reference object whose correlation meets the first requirement is obtained from a plurality of preset reference objects.

[0249] Furthermore, the multimodal information of the target object includes first text information and first image information of the target object; the at least two vectors representing the target object include: a first text vector obtained by vectorizing the first text information and a first image vector obtained by vectorizing the first image information;

[0250] The multimodal information of the reference object includes second text information and second image information of the reference object, and the at least two vectors representing the reference object include: a second text vector obtained by vectorizing the second text information and a second image vector obtained by vectorizing the second image information;

[0251] Accordingly, when determining the correlation between the target object and the reference object based on the at least two vectors representing the target object and the at least two vectors representing the reference object, the determination module 12 is specifically configured to:

[0252] A first similarity between the target object and the reference object is determined based on the first text vector and the second text vector; a second similarity between the target object and the reference object is determined based on the first image vector and the second image vector; and a correlation between the target object and the reference object is determined based on the first similarity and the second similarity.

[0253] Furthermore, the determination module 12 is also used to determine the third similarity between the target object and the reference object based on the first text vector and the second image vector; determine the fourth similarity between the target object and the reference object based on the first image vector and the second text vector; when determining the correlation between the target object and the reference object, it is determined based on the first similarity, the second similarity, the third similarity and the fourth similarity.

[0254] Furthermore, when determining at least one target category related to the target object based on the multimodal information of the target object, the determination module 12 is specifically configured to:

[0255] Vectorize the multimodal information of the target object to obtain at least two vectors representing the target object; obtain category information of multiple preset categories, wherein the category information includes vectors representing the preset categories; determine the correlation between the target object and the preset categories based on the at least two vectors representing the target object and the vector representing the preset categories; and obtain at least one target category from the multiple preset categories whose correlation meets the second requirement.

[0256] Furthermore, the information publishing device provided in this embodiment may also include an expansion module and a vectorized representation module. The expansion module is configured to obtain text information of a preset category; generate consultation questions based on the text information; and use the consultation questions to trigger a natural language processing model to expand the text information of the preset category. The vectorized representation module is configured to vectorize the expanded text information of the preset category to obtain a vector representing the preset category.

[0257] Furthermore, the multimodal information of the target object includes first text information and first image information of the target object; the at least two vectors representing the target object include: a first text vector obtained by vectorizing the first text information and a first image vector obtained by vectorizing the first image information; accordingly,

[0258] When determining the correlation between the target object and the preset category based on the at least two vectors representing the target object and the vector representing the preset category, the determination module 12 is specifically configured to:

[0259] Determine a first similarity between the target object and the category based on the first text vector and the vector representing the preset category; determine a second similarity between the target object and the category based on the first image vector and the vector representing the preset category; and determine the correlation between the target object and the preset category based on the first similarity and / or the second similarity.

[0260] Furthermore, when determining the publishing category to which the target object belongs based on the category to which at least one reference object belongs and at least one target category, the determining module 12 is specifically configured to:

[0261] The category to which at least one reference object belongs and at least one target category are merged to obtain a candidate category list without duplicate categories; the categories in the candidate category list are sorted; and the category in the set position in the sorting order is used as the publishing category to which the target object belongs.

[0262] Furthermore, the target objects are products to be published on the e-commerce platform.

[0263] It should be noted here that the information publishing device provided in the above embodiment can implement the technical solution described in the above corresponding method embodiment. The specific implementation principles of the above modules or units can be found in the contents of the above corresponding method embodiment, which will not be repeated here.

[0264] Figure 9 shows a schematic structural diagram of an information category determination device provided by another embodiment of the present disclosure. As shown in Figure 9, the information category determination device includes: a vectorized representation module 21, an acquisition module 22, and a determination module 23. Among them, the vectorized representation module 21 is used to vectorize the multimodal information of the target object to obtain at least two vectors representing the target object. The acquisition module 22 is used to obtain multiple preset vectors, wherein a preset vector represents a category or a reference object. The determination module 23 is used to determine the preset vector related to the target object based on at least two vectors representing the target object and multiple preset vectors; determine the category to which the target object belongs according to the category represented by the preset vector related to the target object or the category to which the reference object belongs.

[0265] Furthermore, the plurality of preset vectors include:

[0266] Vectorizing the multimodal information of the reference object to obtain at least two vectors representing the reference object; and / or

[0267] The text description of the preset category is vectorized to obtain a vector representing the preset category.

[0268] Furthermore, the information category determination device provided in this embodiment may further include an execution module configured to obtain target object attribute information corresponding to the category to which the target object belongs, and perform a publishing operation using the category to which the target object belongs and the target object attribute information as the target object publishing information.

[0269] It should be noted here that the information category determination device provided in the above embodiment can implement the technical solution described in the above corresponding method embodiment. The specific implementation principles of the above modules or units can be found in the contents of the above corresponding method embodiment, which will not be repeated here.

[0270] Another embodiment of the present disclosure provides an information publishing device, which includes an acquisition module and a display module. The acquisition module is used to obtain multimodal information of the target object in response to a user's operation on the target object. The display module is used to display the publishing category to which the target object belongs based on the multimodal information; wherein the publishing category is determined based on the category to which at least one reference object belongs and / or at least one target category, and there is a correlation between the at least one reference object and / or at least one target category and the multimodal information; the target object attribute information corresponding to the publishing category is displayed for verification, and the publishing operation for the target object is executed after the verification is passed.

[0271] Furthermore, when the display module displays the publishing category to which the target object belongs based on the multimodal information, it is specifically used to:

[0272] Obtain object information of multiple preset reference objects and category information of multiple preset categories; obtain at least one reference object related to the target object from the multiple reference objects based on the multimodal information and the object information of the multiple reference objects; obtain at least one target category related to the target object from the multiple preset categories based on the multimodal information; determine the publication category to which the target object belongs based on the category to which the at least one reference object belongs and the at least one target category.

[0273] It should be noted here that the information publishing device provided in the above embodiment can implement the technical solution described in the above corresponding method embodiment. The specific implementation principles of the above modules or units can be found in the contents of the above corresponding method embodiment, which will not be repeated here.

[0274] The present disclosure also provides an information publishing device according to an embodiment, which includes an acquisition module, a determination module, an execution module, and a display module. The acquisition module is used to obtain multimodal information of the target object in response to the user's operation on the target object; the determination module is used to determine the publishing category to which the target object belongs based on the multimodal information; the publishing category is determined based on the category to which at least one reference object belongs and / or at least one target category, and at least one reference object and / or at least one target category is correlated with the multimodal information. The execution module is used to obtain the target object attribute information corresponding to the publishing category; and execute the publishing operation for the target object according to the publishing category and the target object attribute information. The display module is used to display a preview page after the target object is published.

[0275] It should be noted here that the information publishing device provided in the above embodiment can implement the technical solution described in the above corresponding method embodiment. The specific implementation principles of the above modules or units can be found in the contents of the above corresponding method embodiment, which will not be repeated here.

[0276] Another embodiment of the present disclosure provides a product release device, which includes: an acquisition module, a determination module, an execution module and a display module; wherein the acquisition module is used to obtain multimodal information of the product in response to the merchant's operation on the product; the determination module is used to determine the release category to which the product belongs based on the multimodal information; wherein the release category is determined based on at least one reference product category and / or at least one target category, and the at least one reference product and / or at least one target category is correlated with the multimodal information; and the acquisition module is further used to obtain product attribute information corresponding to the release category; the execution module is used to execute the release operation for the product according to the release category and the product attribute information; and the display module is used to display a preview page after the product is released.

[0277] It should be noted here that the product release device provided in the above embodiment can implement the technical solutions described in the above corresponding method embodiments. The specific implementation principles of the above modules or units can be found in the contents of the above corresponding method embodiments, which will not be repeated here.

[0278] Another embodiment of the present disclosure provides a device for batch publishing of goods. The device comprises: an acquisition module, a determination module, and an execution module. The acquisition module is configured to obtain multimodal information of multiple goods to be batch published in response to a merchant triggering a batch publishing operation. The determination module is configured to determine corresponding publishing information for each of the multiple goods. The execution module is configured to execute the batch publishing operation, so that merchants can view preview pages of multiple goods after they are published.

[0279] The determination module, when determining corresponding release information for one of the multiple products, is specifically configured to:

[0280] Determining a publication category for the product based on the product's multimodal information; wherein the publication category is determined based on a category to which at least one reference object belongs and / or at least one target category, and the at least one reference object and / or at least one target category are correlated with the product's multimodal information;

[0281] According to the multimodal information of the product, the corresponding product attribute information is mounted for the release category to which the product belongs to obtain the release information of the product.

[0282] Furthermore, when the acquisition module responds to the merchant's operation of triggering the batch release of goods and obtains the multimodal information of multiple goods to be released in batches, it is specifically used to: display the batch release page; respond to the merchant's input operation on the batch release page, obtain the batch release file input by the merchant; and read the multimodal information of multiple goods from the batch release file.

[0283] Furthermore, the plurality of products include products of different categories; the multimodal information of the products includes product category information. Accordingly, the product batch publishing device further includes a grouping module. The grouping module is configured to group products with the same product category information. Accordingly, when determining the publishing category to which a product belongs based on the multimodal information of the products, the determination module determines that the publishing category to which one product in the group belongs is the same as the publishing category to which the other products in the same group belong.

[0284] It should be noted here that the commodity batch release device provided in the above embodiment can implement the technical solution described in the above corresponding method embodiment. The specific implementation principles of the above modules or units can be found in the contents of the above corresponding method embodiment, which will not be repeated here.

[0285] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units. That is, they may be located in one place or distributed across multiple network units. Some or all of the modules may be selected based on actual needs to achieve the objectives of the present embodiment. Persons of ordinary skill in the art will be able to understand and implement the present invention without inventive effort.

[0286] Figure 10 shows a schematic diagram of the principle structure of an electronic device provided in accordance with an embodiment of the present disclosure. The electronic device includes a processor 32 and a memory 31. The memory 31 is configured to store one or more computer programs. The processor 32 is coupled to the memory 31 and is configured to execute at least one or more computer programs for implementing the steps of the methods provided in various embodiments of the present disclosure.

[0287] The above-mentioned memory 31 can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk or optical disk.

[0288] Furthermore, as shown in Figure 10, the electronic device further includes other components such as a communication component 33, a power component 34, and an audio component 35. Figure 10 schematically shows only some components, which does not mean that the electronic device only includes the components shown in Figure 10.

[0289] Accordingly, an embodiment of the present disclosure further provides a computer-readable storage medium having computer-executable instructions stored thereon. When the computer-executable instructions are executed by a computer, the method steps or functions provided in the above-mentioned embodiments of the present disclosure can be implemented.

[0290] An embodiment of the present disclosure further provides a computer program product, including a computer program, which implements the method in any of the above method embodiments when the computer program is executed by a processor.

[0291] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present disclosure, rather than to limit them. Although the present disclosure has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present disclosure.

Claims

1. A method for publishing information, wherein: include: Based on the multimodal information of the target object, acquiring at least one reference object related to the target object; Determining at least one target category related to the target object based on the multimodal information of the target object; Determining the publishing category to which the target object belongs according to the category to which the at least one reference object belongs and the at least one target category; Obtaining target object attribute information corresponding to the publishing category; The publishing category and the target object attribute information are used as the publishing information of the target object, and a publishing operation is performed.

2. The method according to claim 1, wherein: Acquiring at least one reference object related to the target object based on multimodal information of the target object includes: Vectorizing the multimodal information of the target object to obtain at least two vectors representing the target object; Acquire object information of a plurality of preset reference objects, wherein the object information of the reference objects includes at least two vectors representing the reference objects; Determining the correlation between the target object and the reference object based on at least two vectors representing the target object and at least two vectors representing the reference object; At least one reference object whose correlation meets the first requirement is obtained from a plurality of preset reference objects.

3. The method according to claim 2, wherein: Vectorizing the multimodal information of the target object, including: The multimodal information of the target object is used as an input of a computing model, and the computing model is executed to output at least two vectors representing the target object.

4. The method according to claim 2, wherein: The multimodal information of the target object includes first text information and first image information of the target object; the at least two vectors representing the target object include: a first text vector obtained by vectorizing the first text information and a first image vector obtained by vectorizing the first image information; The multimodal information of the reference object includes second text information and second image information of the reference object, and the at least two vectors representing the reference object include: a second text vector obtained by vectorizing the second text information and a second image vector obtained by vectorizing the second image information; and Determining the correlation between the target object and the reference object based on at least two vectors representing the target object and at least two vectors representing the reference object includes: Determining a first similarity between the target object and a reference object according to the first text vector and the second text vector; determining a second similarity between the target object and a reference object according to the first image vector and the second image vector; The correlation between the target object and the reference object is determined according to the first similarity and the second similarity.

5. The method according to claim 4, wherein: Also includes: determining a third similarity between the target object and a reference object according to the first text vector and the second image vector; determining a fourth similarity between the target object and a reference object according to the first image vector and the second text vector; When determining the correlation between the target object and the reference object, it is determined according to the first similarity, the second similarity, the third similarity and the fourth similarity.

6. The method according to any one of claims 1 to 5, wherein: Determining at least one target category related to the target object based on the multimodal information of the target object includes: The multimodal information of the target object is vectorized to obtain at least two vectors; Acquire category information of a plurality of preset categories, wherein the category information includes vectors representing the preset categories; Determining the correlation between the target object and the preset category based on at least two vectors representing the target object and a vector representing the preset category; At least one target category whose relevance meets the second requirement is obtained from the multiple preset categories.

7. The method according to claim 6, wherein: Also includes: Get the text information of the preset category; Based on the text information, generate consulting questions; Using the consulting question, triggering a natural language processing model to expand the text information of the preset category; The expanded text information of the preset category is vectorized to obtain a vector representing the preset category.

8. The method according to claim 6, wherein: The multimodal information of the target object includes first text information and first image information of the target object; the at least two vectors representing the target object include: a first text vector obtained by vectorizing the first text information and a first image vector obtained by vectorizing the first image information; and Determining the correlation between the target object and the preset category based on at least two vectors representing the target object and a vector representing the preset category includes: Determining a first similarity between the target object and the category according to the first text vector and the vector representing the preset category; Determining a second similarity between the target object and the category according to the first image vector and the vector representing the preset category; Based on the first similarity and / or the second similarity, the relevance between the target object and a preset category is determined.

9. The method according to any one of claims 1 to 8, wherein: Determining the publishing category to which the target object belongs according to the category to which the at least one reference object belongs and the at least one target category includes: Merging the category to which the at least one reference object belongs and the at least one target category to obtain a candidate category list without duplicate categories; sorting the categories in the candidate category table; The category in the set position in the arrangement order is used as the publishing category to which the target object belongs.

10. The method according to any one of claims 1 to 9, wherein: The target object is the product to be published on the e-commerce platform.

11. A method for determining information categories, wherein: include: Vectorizing the multimodal information of the target object to obtain at least two vectors representing the target object; Acquire multiple preset vectors, wherein one preset vector represents a category or a reference object; Determining a preset vector related to the target object based on the at least two vectors representing the target object and the plurality of preset vectors; The category to which the target object belongs is determined according to the category represented by a preset vector related to the target object or the category to which a reference object belongs.

12. The method according to claim 11, wherein: The plurality of preset vectors include: Vectorizing the multimodal information of the reference object to obtain at least two vectors representing the reference object; and / or The text description of the preset category is vectorized to obtain a vector representing the preset category.

13. The method according to claim 11 or 12, wherein: Also includes: Obtaining target object attribute information corresponding to the category to which the target object belongs; The category to which the target object belongs and the attribute information of the target object are used as the publishing information of the target object, and a publishing operation is performed.

14. A method for publishing information, wherein: include: In response to a user's operation on a target object, acquiring multimodal information of the target object; Based on the multimodal information, displaying the publishing category to which the target object belongs; wherein the publishing category is determined based on the category to which at least one reference object belongs and / or at least one target category, and the at least one reference object and / or the at least one target category are correlated with the multimodal information; The target object attribute information corresponding to the publishing category is displayed for verification, and the publishing operation for the target object is performed after the verification is passed.

15. The method according to claim 14, wherein: Based on the multimodal information, displaying the publishing category to which the target object belongs includes: Obtaining object information of a plurality of preset reference objects and category information of a plurality of preset categories; Acquire at least one reference object related to the target object from the multiple reference objects according to the multimodal information and the object information of the multiple reference objects; Acquire at least one target category related to the target object from the plurality of preset categories according to the multimodal information; Based on the category to which the at least one reference object belongs and the at least one target category, a publishing category to which the target object belongs is determined.

16. A method for releasing a product, wherein: include: In response to a merchant's operation on a product, obtaining multimodal information of the product; Based on the multimodal information, display the publishing category to which the product belongs; wherein the publishing category is determined based on the category to which at least one reference product belongs and / or at least one target category, and the at least one reference product and / or the at least one target category are correlated with the multimodal information; The commodity attribute information corresponding to the publishing category is displayed for verification. After the verification is passed, the publishing operation for the commodity is executed and the commodity is published to the e-commerce platform.

17. A method for publishing information, wherein: include: In response to a user's operation on a target object, acquiring multimodal information of the target object; Based on the multimodal information, determining a publishing category to which the target object belongs; wherein the publishing category is determined based on a category to which at least one reference object belongs and / or at least one target category, and the at least one reference object and / or the at least one target category are correlated with the multimodal information; Obtaining target object attribute information corresponding to the publishing category; Execute a publishing operation for the target object according to the publishing category and the target object attribute information; Display the preview page of the target object after it is published.

18. A method for releasing a product, wherein: include: In response to a merchant's operation on a product, obtaining multimodal information of the product; Based on the multimodal information, determining a publishing category to which the product belongs; wherein the publishing category is determined based on a category to which at least one reference product belongs and / or at least one target category, and the at least one reference product and / or the at least one target category are correlated with the multimodal information; Obtaining commodity attribute information corresponding to the publishing category; Execute a publishing operation for the product according to the publishing category and the product attribute information; Displays a preview page of the product after it is released.

19. A method for batch publishing of commodities, wherein: include: In response to the merchant triggering the batch release of commodities, obtaining multimodal information of multiple commodities to be released in batches; Determine corresponding publishing information for the multiple commodities respectively; Performing a batch publishing operation so that the merchant can view the preview pages of the multiple products in batches after they are published; Wherein, determining corresponding publishing information for one of the multiple commodities includes: Determine the publishing category to which the product belongs based on the multimodal information of the product; wherein the publishing category is determined based on the category to which at least one reference object belongs and / or at least one target category, and the at least one reference object and / or the at least one target category are correlated with the multimodal information of the product; According to the multimodal information of the product, the corresponding product attribute information is mounted for the release category to which the product belongs, so as to obtain Product release information.

20. The method according to claim 19, wherein: In response to the merchant triggering the batch release of commodities, multimodal information of multiple commodities to be released in batches is obtained, including: Display the batch publishing page; In response to an input operation of a merchant on the batch publishing page, obtaining a batch publishing file input by the merchant; The multimodal information of the plurality of commodities is read from the batch release file.

21. The method according to claim 19 or 20, wherein: The multiple commodities include commodities of different categories; The multimodal information of a product contains product category information; as well as The method further comprises: Group products with the same product category information into one group; When determining the publishing category to which a product belongs based on the multimodal information of the product, the publishing category to which one product in a group belongs is determined, and other products in the same group belong to the same publishing category.

22. An information publishing system, wherein: include: The client is used to send multimodal information of the target object to the server in response to the user's operation on the target object; A server, configured to obtain, according to the multimodal information, at least one reference object and / or at least one target category that is relevant to the target object; Determining the publishing category to which the target object belongs according to the category to which the at least one reference object belongs and / or at least one target category; Determine the target object attribute information corresponding to the publishing category; The client is used to display the publishing category and the target object attribute information corresponding to the publishing category for verification; In response to the verification pass instruction, the server is triggered to perform a publishing operation on the target object based on the publishing category and the target object attribute information corresponding to the publishing category.

23. An information publishing system, wherein: include: The client is used to send multimodal information of the target object to the server in response to the user's operation on the target object; The server is configured to obtain, according to the multimodal information of the target object, at least one reference object and / or at least one target category that is relevant to the target object; Determining the publishing category to which the target object belongs according to the category to which the at least one reference object belongs and / or at least one target category; Obtaining target object attribute information corresponding to the publishing category; Execute a publishing operation for the target object according to the publishing category and the target object attribute information; The client is used to display a preview page of the target object after it is published.

24. An electronic device, wherein: include: A memory and a processor, wherein The memory is used to store computer programs; The processor is coupled to the memory and is used to execute the computer program stored in the memory to implement the steps of the information publishing method described in any one of claims 1 to 10, or to implement the steps of the information category determination method described in any one of claims 11 to 13, or to implement the steps of the information publishing method described in claim 14, 15 or 17, or to implement the commodity publishing method described in claim 16 or 18, or to implement the commodity batch publishing method described in any one of claims 19 to 21.

25. A computer-readable storage medium, wherein: The computer-readable storage medium stores computer-executable instructions, which are used to implement the method according to any one of claims 1 to 21 when executed by a processor.

26. A computer program product, comprising a computer program, wherein when the computer program is executed by a processor, the method according to any one of claims 1 to 21 is implemented.

Images