Online commodity size information extraction method and system

Abstract

The invention discloses an online commodity size information extraction method and system. The method comprises the steps of obtaining a dialogue statement of a user in a current man-machine dialogue event; identifying an extraction intention for the target size of the target commodity according to the dialogue statement; in response to the extraction intention, size information of the target size of the target commodity is output according to a commodity marketing page of the target commodity, and a data source of the size information comprises at least one of SKU text information, an SKU main graph, an SKU auxiliary graph and an SPU detail graph in content information of the commodity marketing page. The accuracy, the applicability and the success rate of commodity size information extraction can be improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data processing for business under the Internet industry, or the technical field of electronic commerce service under the new generation of information technology industry, in particular to an online commodity size information extraction method and system. BACKGROUND

[0002] In today's era of rapid development of e-commerce, accurate acquisition of online commodity information is crucial, and commodity size information is one of the key factors for consumers to make decisions. Existing commodity size information extraction technology is usually implemented using large models and simple information extraction strategies. However, the existing commodity size information extraction technology has limited ability to extract three-dimensional size information of commodities, although the picture understanding ability of large models continues to develop. It only stays on the surface understanding of the picture, and it is difficult to accurately obtain key three-dimensional data such as the maximum length and width of the commodity, resulting in inaccurate extraction results and inability to meet actual needs.

[0003] In addition, the existing commodity size information extraction technology does not focus on the extraction of text information and picture information when extracting commodity detail page information, resulting in low extraction efficiency. Moreover, the existing size information extraction technology often has the problem of poor multi-modal information fusion. The existing technology has problems when fusing text, pictures, etc. of the commodity detail page, cannot fully utilize the correlation between text description and picture display, resulting in incomplete or inaccurate information extraction. Moreover, the text and picture information often involve multiple sizes of different models of commodities, and the existing technology cannot accurately extract the size information of the target size of the target model based on data processing technology or electronic commerce service technology, making it difficult to meet the user's demand for accurate and fast acquisition of commodity size information. SUMMARY

[0004] The present application provides an online commodity size information extraction method and system. A terminal device acquires a dialogue sentence of a user in a current human-computer conversation event. According to the dialogue sentence, an extraction intention for a target size of a target commodity is recognized. In response to the extraction intention, size information of the target size of the target commodity is output according to a commodity marketing page of the target commodity. The system can accurately and quickly extract the size information of the target size of the target commodity contained in the commodity marketing page, improving the applicability and success rate of extracting commodity size information.

[0005] In a first aspect, the present application provides an online commodity size information extraction method applied to a terminal device, the method comprising: acquiring a dialogue sentence of a user in a current human-computer conversation event; recognizing an extraction intention for a target size of a target commodity according to the dialogue sentence; In response to the extraction intention, size information of the target size of the target commodity is output according to a commodity marketing page of the target commodity, and data sources of the size information include at least one of the following: SKU text information, SKU main image, SKU auxiliary image, and SPU detail image in content information of the commodity marketing page.

[0006] In a second aspect, the embodiments of the present application provide an online commodity size information extraction system, the system comprising a terminal device and a server in communication connection with the terminal device, wherein The terminal device is configured to execute steps of the method of the first aspect.

[0007] It can be seen that, in the embodiments of the present application, the terminal device acquires a dialogue sentence of a user in a current human-computer dialogue event; identifies an extraction intention for a target size of a target commodity according to the dialogue sentence; and in response to the extraction intention, outputs size information of the target size of the target commodity according to a commodity marketing page of the target commodity, and data sources of the size information include at least one of the following: SKU text information, SKU main image, SKU auxiliary image, and SPU detail image in content information of the commodity marketing page. In this way, compared with the existing commodity size information extraction scheme which does not focus on commodity size information extraction, has poor multi-modal information fusion, and has low extraction model accuracy and applicability, the present application can combine the extraction intention of the user, perform multi-level extraction and judgment operations on the text information and image information of the commodity marketing page, and finally quickly and accurately extract the size information of the target size of the target commodity, thereby improving the accuracy, applicability, and success rate of extracting commodity size information. BRIEF DESCRIPTION OF DRAWINGS

[0008] In order to more clearly illustrate the technical solutions of the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or prior art description. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can also be obtained by those skilled in the art without creative labor.

[0009] Figure 1 is a structural schematic diagram of an online commodity size information extraction system provided by the embodiments of the present application; Figure 2 is a structural schematic diagram of a terminal device provided by the embodiments of the present application; Figure 3 is a step flowchart of an online commodity size information extraction method provided by the embodiments of the present application; Figure 4 is a whole flowchart of an online commodity size information extraction method provided by the embodiments of the present application; Figure 5 is a schematic diagram of a commodity marketing interface provided by an embodiment of the present application; Figure 6 is a schematic diagram of a customer service interface provided by an embodiment of the present application; Figure 7 is a schematic diagram of a commodity information management function interface provided by an embodiment of the present application; Figure 8 is a schematic diagram of a commodity size information extraction application interface provided by an embodiment of the present application; Figure 9 is a schematic diagram of another commodity size information extraction application interface provided by an embodiment of the present application. DETAILED DESCRIPTION

[0010] In order to make the personnel in the art better understand the present application, the technical solutions in the embodiments of the present application will be described clearly and completely below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by the person of ordinary skill in the art without creative labor fall within the scope of protection of the present application.

[0011] The terms "first", "second", and the like in the specification of the present application and the above-described drawings are used to distinguish different objects, rather than to describe a specific order. In addition, the terms "include" and "have" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or device that includes a series of steps or units is not limited to the listed steps or units, but can optionally include steps or units not listed, or can optionally include other steps or units inherent to the process, method, product, or device.

[0012] Reference herein to "an embodiment" means that a particular feature, structure, or characteristic described in connection with the embodiment can be included in at least one embodiment of the present application. The phrase appears at various places in the specification does not necessarily all refer to the same embodiment, nor is it necessarily mutually exclusive of other embodiments. It is explicitly and implicitly understood that the embodiments described herein can be combined.

[0013] In the embodiments of the present application, "and / or" describes the association relationship of the associated objects, which means that there can be three relationships. For example, A and / or B can represent the following three cases: A exists alone; A and B exist simultaneously; B exists alone. Wherein, A and B can be singular or plural.

[0014] In the embodiments of this application, the symbol " / " can represent that the front and rear associated objects are in an "or" relationship. In addition, the symbol " / " can also represent the division symbol, that is, performing division operation. For example, A / B can represent A divided by B.

[0015] In the embodiments of this application, "at least one" or similar expressions refer to any combination of these items, including any combination of single item or multiple items, refer to one or more, and multiple refers to two or more. For example, at least one of a, b or c can represent the following seven cases: a, b, c, a and b, a and c, b and c, a, b and c. Among them, each of a, b and c can be an element or a set containing one or more elements.

[0016] In the embodiments of this application, "equal to" can be used with greater than, which is applicable to the technical solutions adopted when greater than, or can be used with less than, which is applicable to the technical solutions adopted when less than. When equal to is used with greater than, it is not used with less than; when equal to is used with less than, it is not used with greater than.

[0017] The existing commodity size information extraction technology is usually implemented by using a large model and a simple information extraction strategy. However, the existing commodity size information extraction technology has limited extraction capability in the application of a large model, and only stays on the surface understanding of the picture, and it is difficult to accurately obtain key three-dimensional data such as the maximum length and width of the commodity; and the existing commodity size information extraction technology usually does not focus on the extraction of text information and picture information, resulting in low extraction efficiency; and the existing technology often has the problem of poor multi-modal information fusion, which cannot fully utilize the association between text description and picture display, resulting in incomplete or inaccurate information extraction, and thus the existing technology cannot accurately extract the size information of the target size of the target model, and it is difficult to meet the user's accurate and fast acquisition demand for commodity size information.

[0018] In view of the above problems, the embodiments of the present application provide an online commodity size information extraction method and system. The embodiments of the present application will be described in detail below with reference to the accompanying drawings.

[0019] Please refer to Figure 1 , Figure 1 is a structural schematic diagram of an online commodity size information extraction system provided by the embodiments of the present application, as Figure 1 shown, the online commodity size information extraction system 100 includes a server 110 and a terminal device 120.

[0020] The terminal device 120 performs the main operation of size information extraction, and the server 110 is in communication connection with the terminal device 120, and the two work cooperatively. In the extraction of the size information of the commodity, the terminal device 120 receives the user input, processes by using the local model and algorithm, and if the data needs to be supplemented or further confirmed, the terminal device 120 can interact with the server 110, such as requesting the detailed information of the commodity from the server 110, and the server 110 provides support for the terminal device 120.

[0021] Specifically, the terminal device 120 is configured to obtain a dialogue sentence input by a user, and call a pre-trained intent recognition model to recognize an extraction intent for a target size of a target commodity. The terminal device 120 is further configured to obtain size information of the target commodity from a commodity marketing page of the target commodity according to the recognized extraction intent, and the data sources include SKU text information, a SKU main image, a SKU auxiliary image, and an SPU detail image. The terminal device 120 is further configured to output a suggested question sentence and an inquiry channel to prompt the user to obtain the size information through the channel, and receive size information in a question answer result, or obtain an online communication channel interface of the target commodity, send a size information request message to a server device, and then receive and display size information in a size information response message.

[0022] Please refer to Figure 2 , Figure 2 is a structural block diagram of a terminal device provided by an embodiment of the present application, which is configured to execute an online commodity size information extraction system in Figure 1 , as shown in Figure 2 , the terminal device 120 can include one or more of the following components: a memory 23, a processor 21, a communication bus 30, a communication interface 22, and one or more programs 231, the one or more programs 231 are stored in the memory 23 and configured to be executed by the processor 21, and the one or more programs 231 include instructions for executing any step in the following method embodiments. In a specific implementation, the processor 21 is configured to execute any step in the following method embodiments, and when performing data transmission such as sending, the communication interface 22 can be optionally called to complete the corresponding operation. The terminal device 120 can be a mobile phone terminal, a tablet computer, a notebook computer, and a wearable smart device.

[0023] Please refer to Figure 3 , Figure 3 is a step flowchart of an online commodity size information extraction method provided by an embodiment of the present application, which is applied to a terminal device 120 in Figure 1 , as shown in Figure 3 , the method includes the following steps: Step S301, obtaining a dialogue sentence input by a user in a current human-computer dialogue event.

[0024] In a possible implementation, the current human-computer dialogue event is a dialogue event in a customer service interface of a target merchant to which the target product belongs; or The current human-computer dialogue event is a dialogue event in a function interface of a background product information management system of a target merchant to which the target product belongs.

[0025] In a possible implementation, the current human-computer dialogue event can also be a dialogue event for the target product in a target product size information extraction application.

[0026] The target product size information extraction application is a software program for obtaining three-dimensional size data of a product, and is used to solve the demand of a user for product size information in a shopping, design, production and the like, and to improve the information acquisition efficiency and accuracy. In addition, the target product size information extraction application in the present application includes but is not limited to an APP form, and can also exist in the form of a small program, a webpage and the like.

[0027] In a possible implementation, the obtaining of the dialogue statement of the user in the current human-computer dialogue event includes: detecting a first preset operation of the user, the first preset operation including inputting source data corresponding to the target product in a customer service interface of the target merchant, or inputting the source data corresponding to the target product in a function interface of a background product information management system of the target merchant, or inputting the source data corresponding to the target product in the target product size information extraction application, the source data including a product link; in response to the first preset operation, outputting first prompt information, the first prompt information being used to prompt the user to input a target question for the target product corresponding to the source data; detecting that the user inputs a target question for the target product, and determining the dialogue statement of the user in the current human-computer dialogue event based on the first extraction information and the target question.

[0028] The source data corresponding to the target product can also include product images, product videos, product description contents and the like in various presentation modes, and the target product corresponding to the target model can be determined through the source data.

[0029] Exemplarily, the user can complete the dialogue event in the customer service interface of the target merchant to which the target commodity belongs. For example, the user sees a wooden desk on a well-known e-commerce platform and intends to know the detailed size of the wooden desk. Then, the user enters the customer service interface of the store to which the wooden desk belongs and sends a message to the customer service: "Hello, I want to ask the length, width and height of the wooden desk with model X123." The communication between the user and the customer service in this interface is the current human-computer dialogue event.

[0030] Exemplarily, the user can complete the dialogue event in the function interface of the background commodity information management system of the target merchant to which the target commodity belongs. At this time, the user is usually a merchant, and this situation usually occurs in the process of the merchant managing and updating the commodity information. For example, the staff of a merchant selling outdoor tents enters the information of a newly arrived tent (model Y123) in the background commodity information management system. During the input process, the system pops up a prompt box asking: "Please confirm whether the length, width and height of the unfolded tent are correctly input?" Then, the staff replies: "I have checked it. The length is 3 meters, the width is 2 meters, and the height is 1.8 meters." In this process, the interaction between the staff and the system in the function interface of the background commodity information management system is the current human-computer dialogue event.

[0031] Exemplarily, the user can complete the dialogue event in the target commodity size information extraction application for the target commodity. For example, the user installs the target commodity size information extraction application, and the user wants to know the size of the latest smart watch (model Z123). Then, the user opens the application, inputs the link of the smart watch into the dialogue window, and inputs "Please help me check the length, width and thickness of this smart watch." In this application, the query operation initiated by the user constitutes the current human-computer dialogue event.

[0032] In one possible embodiment, the method further comprises: detecting that the user inputs a commodity size keyword in the search box of the e-commerce platform, the commodity size keyword being related to a target size of a target commodity; automatically popping up a reference inquiry question in the search box of the e-commerce platform, the reference inquiry question being used to inquire the size information of the target size of the target commodity; detecting that the user selects the reference inquiry question, and determining the dialogue sentence of the user in the current human-computer dialogue event based on the commodity size keyword and the reference inquiry question.

[0033] Exemplarily, when a user inputs a keyword related to the size of a commodity in a search box of an e-commerce platform, the system can associate some common questions based on the user input. For example, the user inputs "sofa size", and the search box automatically pops up "what is the length, width and height of this sofa". If the user selects this question, the system obtains this pair of dialogue sentences, and can help the user more efficiently obtain the size information of the commodity, and can also guide the user to discover possible ignored questions related to the size of the commodity, thereby improving the accuracy of commodity search and shopping efficiency.

[0034] It can be understood that the current human-computer dialogue event can exist in various dialogue forms, including but not limited to dialogue sentences, pop-up box content selection, etc., and the current human-computer dialogue event can be applied to various scenes, including but not limited to being applied to a customer service interface, a functional interface of a background commodity information management system, a target commodity size information extraction application, or even just being a component existing in an e-commerce platform interface, without forming a direct dialogue to complete the human-computer dialogue event. The existence form of the human-computer dialogue event includes but is not limited to the above-mentioned several examples.

[0035] In step S302, an extraction intention of a target size of a target commodity is recognized according to the dialogue sentence.

[0036] In one possible embodiment, the target size includes a first size in at least one size information of a first model in at least one model of the target commodity, the first size is three-dimensional size information, and the three-dimensional size information is a total length, a total width and a total height of the target commodity of the first model. The recognizing of the extraction intention of the target size of the target commodity according to the dialogue sentence includes: calling a pre-trained intention recognition model to process the dialogue sentence to obtain an extraction intention of the total length, the total width and the total height of the target commodity of the first model.

[0037] It can be understood that the target commodity can include various models, and each model can have various sizes. For example, the target commodity is a sofa, and various models can include a double sofa, a three-person sofa, a ten-person sofa, etc., and various sizes exist for each model of the sofa. For example, in terms of width, there are total width and sitting width. Therefore, based on the pre-trained intention recognition model, the processing can be refined to predict whether the user is asking for the total width or the sitting width according to the dialogue sentence. For example, the dialogue sentence can include the user's question "how many people can the sofa seat, and is it crowded or not", and it is predicted that the user is asking for the sitting width, and the answer contains the sitting width information.

[0038] The intent recognition model can be trained based on a traditional machine learning algorithm or a deep learning algorithm. Specifically, the traditional machine learning algorithm can include a naive Bayes algorithm, that is, based on Bayes theorem and feature conditional independence assumption, the probability of each intent category under a given dialogue sentence is calculated, and the category with the maximum probability is selected as the prediction result; and can include a support vector machine (SVM) algorithm, that is, by finding an optimal hyperplane to separate data of different categories, for linearly inseparable data, the data can also be mapped to a high-dimensional space by a kernel function to make it linearly separable; the deep learning algorithm can include a recurrent neural network (RNN) and its variants (LSTM, GRU), a convolutional neural network (CNN), a Transformer model and the like.

[0039] It can be seen that, in the embodiment, by calling the pre-trained intent recognition model to process the dialogue sentence, the user demand can be accurately analyzed, the three-dimensional size extraction intent for the target product specific model is recognized from the complex and diverse dialogue content, and the accuracy and intelligent degree of information processing are effectively improved, the tediousness and error-prone problem of manual screening and understanding of the dialogue is avoided, key pre-support is provided for subsequent rapid and accurate extraction of product size information, and efficient operation of the entire product size information extraction process is ensured.

[0040] In step S303, in response to the extracted intent, size information of the target size of the target product is output according to a product marketing page of the target product, and the data source of the size information includes at least one of the following: SKU text information, SKU main picture, SKU auxiliary picture and SPU detail picture in the content information of the product marketing page.

[0041] The SKU (Stock Keeping Unit) is the smallest available unit of inventory control. The SKU text information is a detailed textual description of a specific model product, used to distinguish the specific information of different specifications, styles, colors and the like of the same product, and usually includes product name, specification parameter, material, color, target population, size and the like. The SKU main picture is the most important picture of the product in the display list of the e-commerce platform, usually occupies a prominent position, and is the first image seen by consumers. It generally clearly shows the overall appearance, main features and color of the product. The SKU auxiliary picture is a picture that assists the SKU main picture to display other aspects of the product. The number is usually more than the main picture, and the product can be displayed from different angles and different scenes. The SPU is a standardized product unit, which is the smallest unit of product information aggregation. The SPU detail picture is a detailed introduction picture of multiple models of a product, focusing on showing the overall characteristics, functions and use methods of the product, and may include overall appearance display, function demonstration picture, technical parameter description picture, after-sales service commitment and the like.

[0042] In a possible embodiment, the outputting of the size information of the target size of the target commodity according to the commodity marketing page of the target commodity comprises: calling a pre-trained size information extraction model to process content information of the commodity marketing page to obtain the size information of the target size of the target commodity.

[0043] In a possible embodiment, the training process of the size information extraction model comprises importing a training data set and model training; wherein, the training data set comprises a first type of training data set and a second type of training data set, the first type of training data set comprises question and answer data formed by performing multiple first size information question and answers on commodity text information of a commodity of a first category, the second type of training data set comprises question and answer data formed by performing multiple second size information question and answers on a commodity image of the commodity of the first category, and the question contents of the multiple size information question and answers are different; the model training comprises model training of a pre-set first training model according to the first type of training data set and the second type of training data set, and the first training model is pre-trained.

[0044] In a possible embodiment, the training data set further comprises a first target training data set and a second target training data set, the first target training data set is mixed from the first type of training data set and a third type of training data set, the third type of training data set comprises question and answer data formed by performing multiple third size information question and answers on commodity text information of a commodity of a second category, and the second target training data set is mixed from the second type of training data set and a fourth type of training data set, the fourth type of training data set comprises question and answer data formed by performing multiple fourth size information question and answers on a commodity image of the commodity of the second category. The model training further comprises model training of the first training model according to the first target training data set and the second target training data set.

[0045] The commodity text information of the commodity of the first category comprises manually annotated size text information, and the commodity image of the commodity of the first category comprises size image information pre-added by a pre-trained image-text model.

[0046] The first training model can be loaded by using a deep learning framework such as TensorFlow or PyTorch.

[0047] Exemplarily, the first category of goods can be a sofa, and the sofa goods corresponding text information is added with three-dimensional size text information by artificial annotation, and the sofa goods corresponding goods picture is pre-added with three-dimensional size image information by a pre-trained image-text model; further, a plurality of question and answer pairs are created for the sofa corresponding goods text information annotated by artificial, to form a first type of training data set; similarly, a plurality of question and answer pairs are created for the sofa corresponding goods picture, to form a second type of training data set, and the question contents of the plurality of size information questions and answers are different. For example, when the size data of a picture is 100x20x10cm, the following question and answer pairs are obtained, the first question: please help me extract the size of the following picture, and give the answer according to length, width and height, the first answer: [100, 20, 10, cm]; the second question: I want to know the size information of the goods in the picture, can you help me directly extract the size data? The second answer: [100, 20, 10, cm].

[0048] Further, the application adopts an incremental updating method for large model fine-tuning training, which reduces the training time. Specifically, based on the training result each time, for the supplemented data set, such as the data set corresponding to the second category of goods, and the original data set, it is not necessary to retrain, but only to mix the new data set with part of the original data set, so that the training effect of the last batch of data sets can be retained, and the new data set knowledge can be learned faster.

[0049] Further, in the training process of the first training model, the loss function, optimizer and training steps in the training process need to be determined, the model parameters are adjusted according to the loss value through the optimization algorithm, so that the prediction result of the model on a specific task gradually approaches the correct answer marked. Specifically, the cross-entropy loss function is as follows: ; wherein L represents the loss value, N represents the number of samples, i.e. the number of question and answer pairs in the training data, is the true label of the i-th sample (i.e. the correct answer marked, which is the accurate size information in the goods size information extraction task), and the value is usually 0 or 1, is the prediction probability of the model for the i-th sample.

[0050] It can be understood that the cross-entropy loss function measures the deviation degree of the model prediction from the true situation by calculating the logarithmic difference between the true label and the prediction probability and summing up the negative. The goal of model training is to minimize this loss function value, and when the model prediction is closer to the true label, the loss function value is smaller, indicating that the performance of the model on the task is better. In the goods size information extraction, by minimizing the cross-entropy loss, the model can more accurately extract the target size information from the input information.

[0051] In a possible embodiment, the calling the pre-trained size information extraction model to process the content information of the commodity marketing page to obtain the size information of the target size of the target commodity comprises: calling the size information extraction model to perform a first size information extraction operation on the SKU text information in the content information to obtain a first size information extraction result; if it is detected that the first size information extraction result is not empty, determining that the first size information extraction result is the size information of the target size; if it is detected that the first size information extraction result is empty, calling the size information extraction model to perform a second size information extraction operation on the SKU main image and the SKU auxiliary image in the content information to obtain a second size information extraction result; if it is detected that the second size information extraction result is not empty, determining that the second size information extraction result is the size information of the target size; if it is detected that the second size information extraction result is empty, calling the size information extraction model to perform a third size information extraction operation on the SPU detail image in the content information to obtain a third size information extraction result; determining that the third size information extraction result is the size information of the target size.

[0052] Specifically, the above examples are all in the case that the size information is certainly present in the content information of the default commodity marketing page. In actual application scenarios, there can be a case that no size information is present in the content information.

[0053] In a possible embodiment, the method further comprises: if it is detected that the third size information extraction result is empty, outputting a suggested question sentence and an inquiry channel to prompt a user to output the suggested question sentence through the inquiry channel, and receiving and displaying size information in a question reply result; or if it is detected that the third size information extraction result is empty, obtaining an online communication channel interface of the target commodity, and sending the size information request message to a server device through the online communication channel interface, and receiving and displaying size information in a size information response message.

[0054] In a possible embodiment, the calling the size information extraction model to perform a first size information extraction operation on the SKU text information in the content information to obtain a first size information extraction result comprises: performing a text semantic extraction operation on text content of each of a plurality of text regions of the SKU text information by invoking the size information extraction model, to obtain a plurality of text semantic extraction results corresponding to the plurality of text regions, the plurality of text regions including a title text region, a SKU attribute text region, and other text regions in addition to the title text region and the SKU attribute text region; performing a fourth size information pre-extraction operation on the title text region and the SKU attribute text region to obtain a fourth size information pre-extraction result, the fourth size information pre-extraction result being empty or a first correspondence relationship between a reference SKU attribute and size information, the reference SKU attribute including at least one of color and model; If it is detected that the fourth size information pre-extraction result is not empty and there is a first target correspondence relationship in the first correspondence relationship that is semantically consistent with a target size of the target product, the first target size information in the first target correspondence relationship is taken as the first size information extraction result; If it is detected that the fourth size information pre-extraction result is not empty and there is no target correspondence relationship in the first correspondence relationship that is semantically consistent with a target size of the target product, or it is detected that the fourth size information pre-extraction result is empty, performing a fifth size information pre-extraction operation on the other text regions to obtain a fifth size information pre-extraction result, the fifth size information pre-extraction result being empty or a second correspondence relationship between the reference SKU attribute and the size information; If it is detected that the fifth size information pre-extraction result is not empty and there is a second target correspondence relationship in the second correspondence relationship that is semantically consistent with a target size of the target product, the second target size information in the second target correspondence relationship is taken as the first size information extraction result; If it is detected that the fifth size information pre-extraction result is not empty and there is no target correspondence relationship in the second correspondence relationship that is semantically consistent with a target size of the target product, or it is detected that the fifth size information pre-extraction result is empty, determining that the first size information extraction result is empty.

[0055] It can be understood that in the product marketing page in the e-commerce field, the SKU text information is divided into different text regions. The title text region usually contains the core name of the product, the key model, and other information that can directly reflect the main characteristics of the product, to briefly summarize the identity of the product; the SKU attribute text region includes a color classification text region and a specification text region; other text regions include a quantity text region, a guarantee text region, and other text regions. The content corresponding to the SKU attribute text region and the title text region usually directly reflects the core information of the product, and is therefore extracted first.

[0056] Exemplarily, for a tent of a model "Outdoor Explorer-A1", the title text area can be "Outdoor Explorer-A1 Tent", the SKU attribute text area can be written as "Color: Olive Green, Model: Outdoor Explorer-A1, Suitable for 3-4 people", and other text areas can be descriptions of the characteristics of the tent, etc. The model extracts the semantic information of the texts in these areas respectively; then, the fourth size information pre-extraction operation is performed on the title text area and the SKU attribute text area, if the first corresponding relationship between the reference SKU attribute (such as color, model) and the size information is extracted, the judgment continues, if it is empty, the next step is entered; then, if there is a corresponding relationship consistent with the target size semantics in the first corresponding relationship (such as the target is to obtain the length, width and height of the tent after unfolding), the corresponding size information is taken as the first size information extraction result; if there is none or the fourth size information pre-extraction result is empty, the fifth size information pre-extraction operation is performed on other text areas, if information such as "This tent is spacious, with a width of up to 2 meters" is extracted from other text areas, forming a second corresponding relationship, and there is content consistent with the target size semantics, it is taken as the first size information extraction result; if none of the above is satisfied, it is determined that the first size information extraction result is empty, and then the overall process continues to extract from other ways such as SKU main graph and SKU auxiliary graph.

[0057] In one possible embodiment, the calling the size information extraction model to perform the second size information extraction operation on the SKU main graph and the SKU auxiliary graph in the content information obtains a second size information extraction result, comprising: calling the size information extraction model to perform a first image semantic extraction operation on the SKU main graph to obtain a first image semantic extraction result; if it is detected that the first image semantic extraction result is not empty, and there is third target size information representing a target size of the target commodity, the third target size information is taken as the second size information extraction result; if it is detected that the first image semantic extraction result is not empty, and there is no third target size information representing a target size of the target commodity, or it is detected that the first image semantic extraction result is empty, the size information extraction model is called to perform a second image semantic extraction operation on the SKU auxiliary graph to obtain a second image semantic extraction result; if it is detected that the second image semantic extraction result is not empty, and there is fourth target size information representing a target size of the target commodity, the fourth target size information is taken as the second size information extraction result; If it is detected that the second image semantic extraction result is not empty, and the fourth target size information representing the target size of the target commodity does not exist, or it is detected that the second image semantic extraction result is empty, it is determined that the second size information extraction result is empty.

[0058] Exemplarily, the system first calls the size information extraction model to perform a first image semantic extraction operation on the SKU main image, attempts to obtain information about the target commodity size therein, and if the extraction result is not empty and contains third target size information representing the target size, directly takes it as the second size information extraction result. For example, the SKU main image of the tent is labeled with "unfolded size: 2.5 meters long, 2 meters wide", so "2.5 meters long, 2 meters wide" will be determined as the second size information extraction result.

[0059] Further, if the extraction result of the SKU main image does not meet the requirements (i.e. the extraction result is not empty but does not have target size information, or the extraction result is empty), a second image semantic extraction operation is performed on the SKU auxiliary image. If the extraction result of the auxiliary image is not empty and has target size information (fourth target size information), it is taken as the second size information extraction result; if it still does not meet the conditions, it is determined that the second size information extraction result is empty, and subsequent attempts will be made to extract from the SPU detail image. For example, the SKU main image does not display size information, but the auxiliary image shows the internal height details of the tent and is labeled with "internal height 1.8 meters", so "1.8 meters" will be determined as the second size information extraction result.

[0060] In one possible embodiment, the calling the size information extraction model to perform a third size information extraction operation on the SPU detail image in the content information obtains a third size information extraction result, comprising: calling the size information extraction model to perform a third image semantic extraction operation on the SPU detail image obtains a third image semantic extraction result; detecting that the third image semantic extraction result is not empty and there is fifth target size information representing the target size of the target commodity; taking the fifth target size information as the second size information extraction result.

[0061] Exemplarily, the system calls the size information extraction model to perform a third image semantic extraction operation on the SPU detail image. When the extraction result is not empty and there is fifth target size information representing the target size, it is taken as the third size information extraction result. For example, in the SPU detail image of the tent, all size parameters of this model of tent are listed in detail in the form of a chart, including 2.5 meters long, 2 meters wide, and 1.8 meters high, and these information will be extracted as the third size information extraction result.

[0062] In a possible embodiment, the first image semantic extraction operation, the second image semantic extraction operation, and the third image semantic extraction operation refer to extraction operations of numerical information of reference size parameters of each model of the target commodity. The reference size parameters include the total length, the total width, and the total height, or at least two of the reference size parameters can calculate the total length, the total width, or the total height.

[0063] In a possible embodiment, the method further includes: If it is detected that the second image semantic extraction result is not empty, and the fourth target size information representing the target size of the target commodity does not exist, or it is detected that the second image semantic extraction result is empty, it is determined that the second size information extraction result is empty.

[0064] In a possible embodiment, the SPU detail graph includes a plurality of SPU sub-detail graphs; and the calling the size information extraction model to perform a third image semantic extraction operation on the SPU detail graph includes: According to the SKU main graph and the SKU text information, a plurality of search keywords are created, and the plurality of search keywords are associated with model information of a first model of the target commodity; According to the plurality of search keywords, a target SPU sub-detail graph is matched from the plurality of SPU sub-detail graphs based on a preset first matching algorithm, and a matching degree of the target SPU sub-detail graph with the SKU main graph is greater than a preset threshold; The size information extraction model is called to perform a size information extraction operation on the target SPU sub-detail graph, and the third image semantic extraction result is obtained.

[0065] It can be understood that by performing size extraction on the target SPU sub-detail graph in the plurality of SPU sub-detail graphs, it is avoided to extract size information from all SPU sub-detail graphs one by one, the most relevant image to the current target commodity model can be quickly located, and the pertinence and efficiency of information extraction are greatly improved. According to different target commodity models, appropriate images can be flexibly selected from a plurality of corresponding SPU sub-detail graphs for information extraction, which can adapt to a large number of different commodity models and various SPU detail graph display forms on an e-commerce platform, and the applicability of the size information extraction system in a complex commodity data environment is enhanced.

[0066] As can be seen, in this embodiment, the terminal device acquires the user's dialogue statement in the current human-computer dialogue event; identifies the extraction intent for the target size of the target product based on the dialogue statement; and, in response to the extraction intent, outputs the size information of the target product's target size based on the product's marketing page. This enables the system to accurately and quickly extract the size information of the target product contained in the product marketing page, improving the applicability and success rate of extracting product size information.

[0067] Please see Figure 4 , Figure 4 This is an overall flowchart of an online product size information extraction method provided in an embodiment of this application, applied to... Figure 1 Terminal device 120, such as Figure 4 As shown, the method includes the following steps: Step S401: Obtain the dialogue statements.

[0068] The dialogue statement refers to the dialogue statement made by the user in the current human-computer dialogue event.

[0069] In one possible embodiment, the current human-computer dialogue event is a dialogue event in the customer service interface of the target merchant to which the target product belongs; or, the current human-computer dialogue event is a dialogue event in the functional interface of the background product information management system of the target merchant to which the target product belongs.

[0070] In one possible embodiment, the current human-computer dialogue event can also be a dialogue event for the target product in the target product size information extraction application.

[0071] Step S402: Identify the extraction intent for the target size of the target product.

[0072] In one possible embodiment, the target size includes a first size from at least one size information of a first model of at least one model of the target product, wherein the first size is three-dimensional size information, and the three-dimensional size information is the total length, total width, and total height of the target product of the first model; the step of identifying the extraction intent for the target size of the target product based on the dialogue statement includes: calling a pre-trained intent recognition model to process the dialogue statement to obtain the extraction intent for the total length, total width, and total height of the target product of the first model.

[0073] It can be understood that by calling the pre-trained intent recognition model to process the dialogue sentence, the user demand can be accurately analyzed, the three-dimensional size extraction intent for the target commodity specific model can be recognized from the complex and diverse dialogue content, and the accuracy and intelligent degree of information processing are effectively improved, the tedious and error-prone problems of manual screening and understanding of dialogue are avoided, key pre-support is provided for subsequent rapid and accurate extraction of commodity size information, and efficient operation of the entire commodity size information extraction process is ensured.

[0074] In step S403, a first size information extraction operation is performed on the SKU text information in the content information of the target commodity marketing page to obtain a first size information extraction result.

[0075] In a possible implementation, the calling the size information extraction model to perform the first size information extraction operation on the SKU text information in the content information to obtain a first size information extraction result includes: calling the size information extraction model to perform a text semantic extraction operation on text content of each text region in the SKU text information to obtain a plurality of text semantic extraction results corresponding to the plurality of text regions, the plurality of text regions including a title text region, a SKU attribute text region, and other text regions except the title text region and the SKU attribute text region; performing a fourth size information pre-extraction operation on the title text region and the SKU attribute text region to obtain a fourth size information pre-extraction result, the fourth size information pre-extraction result being empty or a first corresponding relationship between a reference SKU attribute and size information, the reference SKU attribute including at least one of color and model; if it is detected that the fourth size information pre-extraction result is not empty and there is a first target corresponding relationship in the first corresponding relationship that is semantically consistent with a target size of the target product, taking first target size information in the first target corresponding relationship as the first size information extraction result; if it is detected that the fourth size information pre-extraction result is not empty and there is no target corresponding relationship in the first corresponding relationship that is semantically consistent with a target size of the target product, or it is detected that the fourth size information pre-extraction result is empty, performing a fifth size information pre-extraction operation on the other text regions to obtain a fifth size information pre-extraction result, the fifth size information pre-extraction result being empty or a second corresponding relationship between the reference SKU attribute and the size information; if it is detected that the fifth size information pre-extraction result is not empty and there is a second target corresponding relationship in the second corresponding relationship that is semantically consistent with a target size of the target product, taking second target size information in the second target corresponding relationship as the first size information extraction result; if it is detected that the fifth size information pre-extraction result is not empty and there is no target corresponding relationship in the second corresponding relationship that is semantically consistent with a target size of the target product, or it is detected that the fifth size information pre-extraction result is empty, determining that the first size information extraction result is empty.

[0076] In step S404, it is determined whether the first size information extraction result is empty.

[0077] Specifically, if yes, step S405 is performed; and if no, step S406 is performed.

[0078] In step S405, a second size information extraction operation is performed on the SKU primary image and the SKU auxiliary image in the content information to obtain a second size information extraction result.

[0079] In a possible embodiment, the calling the size information extraction model to perform the second size information extraction operation on the SKU main image and the SKU auxiliary image in the content information to obtain a second size information extraction result comprises: calling the size information extraction model to perform a first image semantic extraction operation on the SKU main image to obtain a first image semantic extraction result; if it is detected that the first image semantic extraction result is not empty and there is third target size information representing a target size of the target commodity, taking the third target size information as the second size information extraction result; if it is detected that the first image semantic extraction result is not empty and there is no third target size information representing a target size of the target commodity, or it is detected that the first image semantic extraction result is empty, calling the size information extraction model to perform a second image semantic extraction operation on the SKU auxiliary image to obtain a second image semantic extraction result; if it is detected that the second image semantic extraction result is not empty and there is fourth target size information representing a target size of the target commodity, taking the fourth target size information as the second size information extraction result; if it is detected that the second image semantic extraction result is not empty and there is no fourth target size information representing a target size of the target commodity, or it is detected that the second image semantic extraction result is empty, determining that the second size information extraction result is empty.

[0080] In step S406, it is determined that the first size information extraction result is size information of a target size.

[0081] In step S407, it is determined whether the second size information extraction result is empty.

[0082] Specifically, if yes, step S408 is performed; and if no, step S409 is performed.

[0083] In step S408, a third size information extraction operation is performed on the SPU detail image in the content information to obtain a third size information extraction result.

[0084] In a possible embodiment, the calling the size information extraction model to perform the third size information extraction operation on the SPU detail image in the content information to obtain a third size information extraction result comprises: calling the size information extraction model to perform a third image semantic extraction operation on the SPU detail image to obtain a third image semantic extraction result; detecting that the third image semantic extraction result is not empty and there is fifth target size information representing a target size of the target commodity; and taking the fifth target size information as the second size information extraction result.

[0085] In a possible embodiment, the first image semantic extraction operation, the second image semantic extraction operation, and the third image semantic extraction operation refer to extraction operations of numerical information of reference size parameters of each model of the target commodity; wherein the reference size parameters include the total length, the total width, and the total height, or at least two of the reference size parameters can calculate the total length, or the total width, or the total height.

[0086] At step S409, a third size information extraction operation is performed on the SPU detail image in the content information to obtain a third size information extraction result.

[0087] At step S410, it is determined that the third size information extraction result is the size information of the target size.

[0088] In a possible embodiment, the method further includes: If it is detected that the third size information extraction result is empty, a suggestion question sentence and an inquiry channel are output to prompt the user to output the suggestion question sentence through the inquiry channel; and size information in a question reply result is received and displayed; or, If it is detected that the third size information extraction result is empty, an online communication channel interface of the target commodity is obtained, and the size information request message is sent to the server device through the online communication channel interface, and size information in a size information response message is received and displayed.

[0089] It can be seen that in the embodiment, the terminal device obtains a dialogue sentence of the user in a current human-computer dialogue event; identifies an extraction intention for a target size of a target commodity according to the dialogue sentence; and in response to the extraction intention, outputs size information of the target size of the target commodity according to a commodity marketing page of the target commodity, so that the system has an accurate and fast extraction capability for size information of the target size of the target commodity contained in the commodity marketing page, and the applicability and success rate of extracting the size information of the commodity are improved.

[0090] Please refer to Figure 5 , Figure 5 is a schematic diagram of a commodity marketing interface provided by an embodiment of the present application, as Figure 5 shown, the content information of the commodity marketing interface includes SKU text information, a SKU main image, a SKU auxiliary image, and an SPU detail image.

[0091] The SKU text information includes a plurality of text areas, specifically including a title text area, a SKU attribute text area, and other text areas, wherein the SKU attribute text area includes a color classification text area and a specification text area, and the other text areas include a delivery text area, a guarantee text area, a quantity text area, and the like.

[0092] Specifically, as shown in Figure 5 the upper left region of the interface shows a picture of a three-seater sofa, and the current main SKU picture can be switched to an auxiliary SKU picture through the left and right switching buttons below the picture; and the right region of the main SKU picture includes a title text region, a SKU attribute text region, a color classification text region, a delivery text region, a guarantee text region, and a quantity text region, which respectively present the title information as "leisure sofa", color classification information, delivery information, guarantee information, quantity information, and purchase information.

[0093] Further, the lower middle region of the interface is a product detail region. As can be seen, the product detail region displays pictures and attribute information of different models corresponding to the sofa product, specifically including a two-seater model and a three-seater model, and the attribute information includes fabric information, filling information, frame material information, etc. Among them, the pictures of the two-seater sofa and the three-seater sofa are marked with size information of multiple sizes.

[0094] It can be understood that a product marketing page usually includes SKU text information, a main SKU picture, an auxiliary SKU picture, and an SPU detail picture. Among them, the main SKU picture is used to present one model information of a product, and the SPU detail picture is used to present multiple model information of a product. Therefore, the content information of the existing product marketing page is relatively rich and diverse. If only the existing product size information extraction technology is used, it is difficult to implement a multi-level extraction strategy for the size information of the target model of the target product based on the content information in the product marketing page, and it is difficult to meet the user's demand for accurate and fast acquisition of product size information.

[0095] Please refer to Figure 6 , Figure 6 is a schematic diagram of a customer service interface provided by an embodiment of the present application, as shown in Figure 6 the customer service interface is a customer service interface of a target merchant to which a target product belongs, and the customer service interface presents an application scenario of product size information extraction.

[0096] In the customer service interface, there is a man-machine dialogue event, the machine customer service or the artificial customer service asks the user "Hello, what's your problem?", and then the user side sends a target commodity link and a picture of the target commodity, and asks "What is the size of this sofa? How many people can it seat?" The customer service replies "The three-dimensional size of this sofa is 157, 60, 71 cm, among which the seat width is 146 cm and it can seat three people." It can be understood that when the user asks the size information of the target commodity in the customer service interface, the relevant information of the target commodity needs to be clear, for example, the target commodity link, the commodity picture, or the description content of the specific model commodity can be sent to make the terminal device clear the extraction intention of the target size of the target commodity, and then complete a dialogue sentence and output the final size extraction result from the terminal device side. In this way, the present application can meet the accurate and fast acquisition demand of the user for the commodity size information.

[0097] Please refer to Figure 7 , Figure 7 is a schematic diagram of a commodity information management function interface provided by an embodiment of the present application, as Figure 7 indicated, the commodity information management function interface is a function interface of a background commodity information management system of a target merchant to which the target commodity belongs, and the commodity information management function interface presents another application scenario of commodity size information extraction.

[0098] Among them, the "commodity display area" on the left side of the interface is used to display the related information of the user-entered to-be-processed commodities, such as commodity 1, commodity 2, and commodity 3; the "commodity information management area" on the right side of the interface is a key area for size information extraction and entry. For the to-be-processed commodities, the size information can be obtained and managed through this area, which is an operation entrance for the terminal device to process the commodity size information. Specifically, in this area, multiple information of the to-be-processed commodities can be edited, including but not limited to the commodity name, the commodity model, and the commodity size. Among them, the editing of the commodity size can realize the extraction of the commodity size through manual editing, one-key acquisition, intelligent extraction, and other ways, and then the size information management is carried out.

[0099] Specifically, the user can manually input the product size in the text box after "length", "width", "height", which is a manual supplement method; the one-key acquisition function is used to determine the three-dimensional size information corresponding to the size type after selecting the overall size of the sofa, for example, based on the online product size information extraction method provided in the present application, and automatically fill in the text box; intelligent extraction is used for the user's more complex product size acquisition requirements, based on real-time dialogue, the user's extraction intention for the size information of the to-be-processed product is determined, and then the size information required by the user is output based on the online product size information extraction method provided in the present application. For example, the user side needs to obtain "the length and width of the product of the first size, and the height of the product of the second size", so it needs to involve multiple dimensions of size information corresponding to two types of models, and the one-key acquisition function cannot obtain multiple types of size information at one time, and then the intelligent dialogue is needed to quickly and accurately obtain the size information.

[0100] It can be understood that this interface is mainly applied to the e-commerce platform or the merchant background management scene, which is convenient for unified management and updating of product size information. Through these size information extraction methods, it is ensured that the size information of the product displayed on the front end is accurate and correct, which meets the user's demand for understanding the product specifications, and also provides accurate size data support for subsequent logistics, warehousing and other links.

[0101] Please refer to Figure 8 , Figure 8 is a schematic diagram of a product size information extraction application interface provided by an embodiment of the present application, as Figure 8 indicated, the interface presents another application scenario of product size information extraction.

[0102] Among them, the target product size information extraction application is a software program for acquiring three-dimensional size data of a product, which aims to solve the user's demand for product size information in the scenes of shopping, design, production, etc., and improve the information acquisition efficiency and accuracy. In addition, the target product size information extraction application in the present application includes but is not limited to the form of APP, and can also exist in the form of small programs, web pages and other corresponding forms.

[0103] Among them, the upper half of the interface is the "product information uploading area", which provides multiple data entry methods for the user, which can input product links, pictures or description contents by clicking, or can be uploaded by dragging, to provide original data for subsequent extraction of size information by using models and algorithms; and the lower half of the interface is the "dialogue interface", which is convenient for the user to input the size information related instructions that the user wants to obtain.

[0104] Further, as Figure 9 indicated, Figure 9is another application interface schematic diagram of the commodity size information extraction application provided in the embodiment of the present application. When the user outputs the commodity data in the "commodity information uploading area", the "commodity information display area" presents the basic information such as the commodity name and model, and then the user's understanding of the commodity can be improved; and a complete interactive scene is displayed in the "dialogue interface", the user inquires the size and the number of people that can be seated of the three-seater sofa, and the system makes an accurate reply, which embodies the application process of obtaining and feeding back the size information through dialogue interaction in the scheme, and also shows the processing capability of the application to the user's demand in actual use, that is, according to the input instruction, the related model and data source are called to extract and return the accurate size information.

[0105] It can be seen that in the embodiment, the multi-level extraction and judgment operation is performed on the text information and image information of the commodity marketing page in combination with the extraction intention of the user, and finally the size information of the target size of the target commodity is quickly and accurately extracted. Compared with the existing commodity size information extraction scheme which does not focus on the extraction of commodity size information, the multi-modal information fusion is not good, and the extraction model accuracy and applicability are not high, the present application improves the accuracy, applicability and success rate of extracting commodity size information.

[0106] In addition, the embodiment of the present application also provides a computer storage medium which stores a computer program capable of being loaded and executed by a processor, such as a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and various media which can store program codes.

[0107] It should be noted that, for the foregoing method embodiments, in order to simply describe, they are all expressed as a series of action combinations, but those skilled in the art should know that the present application is not limited to the action sequence described, because according to the present application, certain steps can be performed in other sequences or simultaneously. Secondly, those skilled in the art should know that the embodiments described in the specification all belong to preferred embodiments, and the actions and units involved are not necessarily necessary for the present application.

[0108] It is only a logical function division, and actual implementation can have another division mode; for example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the coupling or direct coupling or communication connection between the units or components shown or discussed can be indirect coupling or communication connection through some interfaces, devices or units, which can be electrical, mechanical or other forms.

[0109] The units described as separate components may or may not be physically separate, and the components displayed as units may or may not be physical units, i.e. may be located in one place, or may be distributed on multiple network units. Part or all of the units may be selected according to actual needs to achieve the purpose of the embodiment.

[0110] In addition, the functional units in various embodiments of the application can be integrated in one processing unit, or each unit can be physically included separately, or two or more units can be integrated in one unit. The integrated unit can be realized in the form of hardware, or in the form of hardware plus software functional units.

[0111] The integrated unit realized in the form of software functional units can be stored in a computer readable storage medium. The software functional units stored in a storage medium include a plurality of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute part of the steps of the method described in various embodiments of the application. The foregoing storage medium includes a U disk, a mobile hard disk, a magnetic disk, an optical disk, a volatile memory or a non-volatile memory. The non-volatile memory can be a read-only memory (read-only memory, ROM), a programmable read-only memory (programmable ROM, PROM), an erasable programmable read-only memory (erasable PROM, EPROM), an electrically erasable programmable read-only memory (electrically EPROM, EEPROM) or a flash memory. The volatile memory can be a random access memory (random access memory, RAM) used as an external cache. By way of example but not limitation, many forms of random access memory (random access memory, RAM) can be used, such as static random access memory (static RAM, SRAM), dynamic random access memory (DRAM), synchronous dynamic random access memory (synchronous DRAM, SDRAM), double data rate synchronous dynamic random access memory (double data rate SDRAM, DDR SDRAM), enhanced synchronous dynamic random access memory (enhanced SDRAM, ESDRAM), synchronous link dynamic random access memory (synch link DRAM, SLDRAM) and direct memory bus random access memory (direct rambus RAM, DRRAM) and various media that can store program codes.

[0112] In the above embodiments, the description of each embodiment is focused on, and the part not described in detail in a certain embodiment can be referred to the relevant description of other embodiments.

[0113] The above has introduced the embodiments of the present application in detail, and the principle and implementation mode of the present application are described by applying specific examples; the above embodiment explanation is only for helping to understand the method of the present application and its core idea; meanwhile, for the general technical personnel in the art, according to the idea of the present application, the specific implementation mode and application range can be changed, and the above-mentioned description should not be understood as the limitation of the present application.

[0114] Although the present application is disclosed as above, the present application is not limited to this. Any person skilled in the art can easily think of changes or substitutions without departing from the spirit and scope of the present application, and can make various changes and modifications, including the combination of different functions and implementation steps, including the software and hardware implementation mode, which are all within the protection scope of the present application.

Claims

1. An online product dimension information extraction method, characterized by, Applied to a terminal device, comprising: obtaining a dialogue sentence of a user in a current human-computer dialogue event; identifying an extraction intention of a target size of a target commodity according to the dialogue sentence; in response to the extraction intention, outputting size information of the target size of the target commodity according to a commodity marketing page of the target commodity, wherein the data source of the size information comprises at least one of the following: SKU text information, SKU main picture, SKU auxiliary picture and SPU detail picture in content information of the commodity marketing page.

2. The method of claim 1, wherein, The target size comprises a first size in at least one size information of a first model in at least one model of the target commodity, and the first size is three-dimensional size information, and the three-dimensional size information is the total length, total width and total height of the target commodity of the first model. The identification of the extraction intention of the target size of the target commodity according to the dialogue sentence comprises: calling a pre-trained intention recognition model to process the dialogue sentence to obtain the extraction intention of the total length, total width and total height of the target commodity of the first model.

3. The method of claim 2, wherein, The outputting of the size information of the target size of the target commodity according to the commodity marketing page of the target commodity comprises: calling a pre-trained size information extraction model to process the content information of the commodity marketing page to obtain the size information of the target size of the target commodity.

4. The method of claim 3, wherein, The calling of the pre-trained size information extraction model to process the content information of the commodity marketing page to obtain the size information of the target size of the target commodity comprises: calling the size information extraction model to perform a first size information extraction operation on the SKU text information in the content information to obtain a first size information extraction result; if it is detected that the first size information extraction result is not empty, determining that the first size information extraction result is the size information of the target size; if it is detected that the first size information extraction result is empty, calling the size information extraction model to perform a second size information extraction operation on the SKU main picture and the SKU auxiliary picture in the content information to obtain a second size information extraction result; if it is detected that the second size information extraction result is not empty, determining that the second size information extraction result is the size information of the target size; if it is detected that the second size information extraction result is empty, calling the size information extraction model to perform a third size information extraction operation on the SPU detail picture in the content information to obtain a third size information extraction result; determining that the third size information extraction result is the size information of the target size.

5. The method of claim 4, wherein, The calling of the size information extraction model to perform a first size information extraction operation on the SKU text information in the content information to obtain a first size information extraction result comprises: The size information extraction model is called to perform a text semantic extraction operation on text content of each of a plurality of text regions of the SKU text information, to obtain a plurality of text semantic extraction results corresponding to the plurality of text regions, the plurality of text regions including a title text region, a SKU attribute text region, and other text regions in addition to the title text region and the SKU attribute text region; A fourth size information pre-extraction operation is performed on the title text region and the SKU attribute text region, to obtain a fourth size information pre-extraction result, the fourth size information pre-extraction result being empty or a first correspondence relationship between a reference SKU attribute and size information, the reference SKU attribute including at least one of color and model; If it is detected that the fourth size information pre-extraction result is not empty and there is a first target correspondence relationship in the first correspondence relationship that is semantically consistent with a target size of the target commodity, first target size information in the first target correspondence relationship is taken as the first size information extraction result; If it is detected that the fourth size information pre-extraction result is not empty and there is no target correspondence relationship in the first correspondence relationship that is semantically consistent with a target size of the target commodity, or it is detected that the fourth size information pre-extraction result is empty, a fifth size information pre-extraction operation is performed on the other text regions, to obtain a fifth size information pre-extraction result, the fifth size information pre-extraction result being empty or a second correspondence relationship between the reference SKU attribute and the size information; If it is detected that the fifth size information pre-extraction result is not empty and there is a second target correspondence relationship in the second correspondence relationship that is semantically consistent with a target size of the target commodity, second target size information in the second target correspondence relationship is taken as the first size information extraction result; If it is detected that the fifth size information pre-extraction result is not empty and there is no target correspondence relationship in the second correspondence relationship that is semantically consistent with a target size of the target commodity, or it is detected that the fifth size information pre-extraction result is empty, it is determined that the first size information extraction result is empty.

6. The method of claim 5, wherein, The calling of the size information extraction model to perform the second size information extraction operation on the SKU main image and the SKU auxiliary image in the content information to obtain the second size information extraction result includes: The size information extraction model is called to perform a first image semantic extraction operation on the SKU main image, to obtain a first image semantic extraction result; If it is detected that the first image semantic extraction result is not empty and there is third target size information representing a target size of the target commodity, the third target size information is taken as the second size information extraction result; If it is detected that the first image semantic extraction result is not empty, and the third target size information representing the target size of the target commodity does not exist, or it is detected that the first image semantic extraction result is empty, the size information extraction model is called to perform a second image semantic extraction operation on the SKU auxiliary image to obtain a second image semantic extraction result; If it is detected that the second image semantic extraction result is not empty, and the fourth target size information representing the target size of the target commodity exists, the fourth target size information is taken as the second size information extraction result; If it is detected that the second image semantic extraction result is not empty, and the fourth target size information representing the target size of the target commodity does not exist, or it is detected that the second image semantic extraction result is empty, it is determined that the second size information extraction result is empty.

7. The method of claim 6, wherein, The calling of the size information extraction model to perform a third size information extraction operation on the SPU detail image in the content information to obtain a third size information extraction result comprises: Calling the size information extraction model to perform a third image semantic extraction operation on the SPU detail image to obtain a third image semantic extraction result; Detecting that the third image semantic extraction result is not empty, and that the fifth target size information representing the target size of the target commodity exists; Taking the fifth target size information as the second size information extraction result.

8. The method of claim 7, wherein, The first image semantic extraction operation, the second image semantic extraction operation, and the third image semantic extraction operation refer to extraction operations of numerical information of reference size parameters of each model of the target commodity; The reference size parameters include the total length, the total width, and the total height, or at least two reference size parameters in the reference size parameters can calculate the total length, or the total width, or the total height.

9. The method according to any one of claims 1 to 8, characterized in that, The current human-computer dialogue event is a dialogue event in a customer service interface of a target merchant to which the target commodity belongs; Or, The current human-computer dialogue event is a dialogue event in a function interface of a background commodity information management system of a target merchant to which the target commodity belongs.

10. An on-line product dimension information extraction system, characterized by, The system comprises a terminal device and a server in communication connection with the terminal device, wherein The terminal device is configured to perform steps in any one of the methods of claims 1-9.

Images