Object feature data generation method, electronic equipment and storage medium

Abstract

The invention discloses an object feature data generation method, electronic equipment and a storage medium, and relates to the technical field of artificial intelligence and natural language processing. The method comprises the following steps: acquiring multi-modal information of a target object; based on the multi-modal information of the target object, at least one candidate object is determined, and the similarity between the data distribution state of the candidate object and the data distribution state of the target object exceeds a first threshold value; feature data of the candidate objects are obtained, and the feature data of the candidate objects are used for representing expression results of the candidate objects on the object recommendation platform; on the basis of the feature data of the candidate objects, feature data of the target object is obtained through prediction, and the feature data of the target object is used for representing the expression result of the target object on the object recommendation platform. The technical problem that the data generation efficiency of the object is low is solved.

Description

Technical Field

[0001] This application relates to the fields of artificial intelligence technology and natural language processing technology. Specifically, it relates to a method for generating feature data of an object, an electronic device, and a storage medium. Background Art

[0002] Currently, on an object recommendation platform, when determining the feature data of an object, usually based on the behavior data of the object, the weights of the feature sequences of the object are determined, and then the feature sequences are weighted and fused through these weights to obtain the feature data of the object.

[0003] However, since an object usually lacks historical behavior data, effective feature data of the object cannot be obtained based on the above method, resulting in the problem of low data generation efficiency of the object.

[0004] For the above problems, no effective solution has been proposed yet. Summary of the Invention

[0005] Embodiments of this application provide a method for generating feature data of an object, an electronic device, and a storage medium to at least solve the technical problem of low data generation efficiency of the object.

[0006] According to one aspect of the embodiments of this application, a method for generating feature data of an object is provided. The method may include: obtaining multi-modal information of a target object; based on the multi-modal information of the target object, determining at least one candidate object, where the similarity between the data distribution state of the candidate object and the data distribution state of the target object exceeds a first threshold; obtaining the feature data of the candidate object, where the feature data of the candidate object is used to represent the performance result of the candidate object on the object recommendation platform; predicting the feature data of the target object based on the feature data of the candidate object, where the feature data of the target object is used to represent the performance result of the target object on the object recommendation platform.

[0007] According to another aspect of the embodiments of this application, a method for generating feature data of an object is provided. The method may include: obtaining multi-modal information of a target trading object; based on the multi-modal information of the target trading object, determining at least one candidate trading object, where the similarity between the data distribution state of the candidate trading object and the data distribution state of the target trading object exceeds a first threshold; obtaining the feature data of the candidate trading object, where the feature data of the candidate trading object is used to represent the performance result of the candidate trading object on the e-commerce platform; predicting the feature data of the target trading object based on the feature data of the candidate trading object, where the feature data of the target trading object is used to represent the performance result of the target trading object on the e-commerce platform.

[0008] According to another aspect of the embodiments of the present application, a method for generating feature data of an object is provided. The method may include: in response to an input operation on an operation interface, displaying multimodal information of a target object on the operation interface; in response to a feature prediction operation on the operation interface, displaying the feature data of the target object, where the feature data of the target object is predicted based on the feature data of at least one candidate object and is used to represent the performance result of the target object on an object recommendation platform, the feature data of the candidate object is used to represent the performance result of the candidate object on the object recommendation platform, the candidate object is determined based on the multimodal information of the target object, and the similarity between the data distribution state of the candidate object and the data distribution state of the target object exceeds a first threshold.

[0009] According to another aspect of the embodiments of the present application, a method for generating feature data of an object is provided. The method may include: by calling a first interface, obtaining multimodal information of a target object, where the first interface includes a first parameter, and the parameter value of the first parameter includes the multimodal information of the target object; based on the multimodal information of the target object, determining at least one candidate object, where the similarity between the data distribution state of the candidate object and the data distribution state of the target object exceeds a first threshold; obtaining the feature data of the candidate object, where the feature data of the candidate object is used to represent the performance result of the candidate object on an object recommendation platform; predicting the feature data of the target object based on the feature data of the candidate object, where the feature data of the target object is used to represent the performance result of the target object on an object recommendation platform; and outputting the feature data of the target object by calling a second interface, where the second interface includes a second parameter, and the parameter value of the second parameter includes the feature data of the target object.

[0010] According to another aspect of the embodiments of the present application, an object feature data generation system is provided. The system may include: a client for transmitting multimodal information of a target object; a server for determining at least one candidate object based on the multimodal information of the target object, where the similarity between the data distribution state of the candidate object and the data distribution state of the target object exceeds a first threshold; obtaining the feature data of the candidate object, where the feature data of the candidate object is used to represent the performance result of the candidate object on an object recommendation platform; predicting the feature data of the target object based on the feature data of the candidate object, where the feature data of the target object is used to represent the performance result of the target object on an object recommendation platform; and returning the feature data of the target object to the client.

[0011] According to another aspect of the embodiments of the present application, a computing device is further provided, including: a memory storing an executable program; a processor for running the program, where when the program runs, it executes the methods in the various embodiments of the present application.

[0012] According to another aspect of the embodiments of the present application, there is also provided an electronic device, including: a memory storing an executable program; a processor connected to the memory through a bus for running the program, wherein when the program runs, it executes the methods in the various embodiments of the present application.

[0013] According to another aspect of the embodiments of the present application, there is also provided a computer-readable storage medium, the computer-readable storage medium including a stored executable program, wherein when the executable program runs, it controls the device where the computer-readable storage medium is located to execute the methods in the various embodiments of the present application.

[0014] According to another aspect of the embodiments of the present application, there is also provided a computer program product, including a computer program, the computer program realizing the methods in the various embodiments of the present application when executed by a processor.

[0015] According to another aspect of the embodiments of the present application, there is also provided a computer program product, including a non-volatile computer-readable storage medium storing a computer program, the computer program realizing the methods in the various embodiments of the present application when executed by a processor.

[0016] According to another aspect of the embodiments of the present application, there is also provided a computer program, the computer program realizing the methods in the various embodiments of the present application when executed by a processor.

[0017] In the embodiments of the present application, multi-modal information of a target object is obtained; based on the multi-modal information of the target object, at least one candidate object is determined, wherein the similarity between the data distribution state of the candidate object and the data distribution state of the target object exceeds a first threshold; feature data of the candidate object is obtained, wherein the feature data of the candidate object is used to represent the performance result of the candidate object on the object recommendation platform; based on the feature data of the candidate object, the feature data of the target object is predicted, wherein the feature data of the target object is used to represent the performance result of the target object on the object recommendation platform. That is to say, in the embodiments of the present application, based on the multi-modal information of the target object, a candidate object consistent with the data distribution state of the target object is determined, and then through the feature data of the candidate object, the feature data for characterizing the performance result of the target object is predicted, avoiding the situation that due to the lack of behavioral data of the target object, the feature data cannot be effectively determined based on the behavioral data of the target object, achieving the purpose of accurately describing the performance result of the target object, and further realizing the technical effect of improving the data generation efficiency of the object, and solving the technical problem of low data generation efficiency of the object.

[0018] It is easy to notice that the above general description and the following detailed description are only for exemplifying and explaining the present application, and do not constitute a limitation to the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] The accompanying drawings described herein are used to provide a further understanding of the present application and form a part of the present application. The schematic embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation of the present application. In the drawings:

[0020] Figure 1 is a schematic diagram of an application scenario of a method for generating feature data of an object according to an embodiment of the present application;

[0021] Figure 2 is a flowchart of a method for generating feature data of an object according to an embodiment of the present application;

[0022] Figure 3 is a flowchart of another method for generating feature data of an object according to an embodiment of the present application;

[0023] Figure 4 is a flowchart of yet another method for generating feature data of an object according to an embodiment of the present application;

[0024] Figure 5 is a flowchart of yet another method for generating feature data of an object according to an embodiment of the present application;

[0025] Figure 6 is a schematic diagram of a system for generating feature data of an object according to an embodiment of the present application;

[0026] Figure 7 is a schematic diagram of generating similar sequence features of new products according to an embodiment of the present application;

[0027] Figure 8 is a schematic diagram of predicting the future performance and maximum potential of new products according to an embodiment of the present application;

[0028] Figure 9 is a schematic diagram of calling a new product potential prediction model according to an embodiment of the present application;

[0029] Figure 10 is a schematic diagram of a device for generating feature data of an object according to an embodiment of the present application;

[0030] Figure 11 is a schematic diagram of another device for generating feature data of an object according to an embodiment of the present application;

[0031] Figure 12 is a schematic diagram of yet another device for generating feature data of an object according to an embodiment of the present application;

[0032] Figure 13 is a schematic diagram of yet another device for generating feature data of an object according to an embodiment of the present application;

[0033] Figure 14 It is a structural block diagram of a computing device according to an embodiment of the present application;

[0034] Figure 15 It is a structural block diagram of an electronic device according to an embodiment of the present application. Detailed implementation manners

[0035] In order to enable those skilled in the art to better understand the solutions of the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only a 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 those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present application.

[0036] It should be noted that the terms "first", "second", etc. in the specification, claims and above-mentioned drawings of the present application are used to distinguish similar objects, and do not necessarily need to be used to describe a specific order or sequence. It should be understood that such data can be interchanged under appropriate circumstances so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein. 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 does not necessarily need to be limited to those steps or units clearly listed, but may include other steps or units not clearly listed or inherent to these processes, methods, products or devices.

[0037] The technical solutions provided by the present application are mainly implemented by using deep learning models. Deep learning models can be widely applied in fields such as natural language processing (NLP), computer vision, and speech processing. Specifically, they can be applied to tasks in the field of computer vision such as visual question answering (VQA), image captioning (IC), and image generation, and can also be widely applied to tasks in the field of natural language processing such as text-based sentiment classification, text summary generation, and machine translation. Therefore, the main application scenarios of the present application include but are not limited to digital assistants, intelligent robots, search, online education, office software, e-commerce, intelligent design, etc.

[0038] First, some nouns or terms that appear in the process of describing the embodiments of the present application are applicable to the following explanations:

[0039] Multimodal, which simultaneously utilizes and processes two or more different types of data or information sources for collaborative reasoning. Multimodal can include, but is not limited to, images, text, audio, video, etc.;

[0040] Image-text relevance matching. Through a contrastive learning method, a pre-trained model can understand the relevance between image content and text descriptions. By establishing multimodal representations of images and text, a cross-modal mapping relationship can be learned, enabling the ability to understand and match images with relevant text descriptions without specific task training;

[0041] The target-attention mechanism is typically applied in recommendation tasks to perform attention operations on target objects and related candidate objects. The target object (e.g., the item to be scored) can be used as a query, and the retrieved candidate objects (e.g., a sequence of similar items) can be used as keys. Using the target-attention method, the similarity weights between the target object and candidate objects can be calculated.

[0042] According to an embodiment of the present application, a method for generating feature data of an object is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. And although the logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in a different order than here.

[0043] The deep learning model involved in the embodiments of the present application can be an artificial intelligence-based language model (Language Mode, LM) or a multimodal model (Multimodal Model, MM), etc. The embodiments of the present application do not limit the number of model parameters supported by the model, aiming to meet actual needs. If the model parameters are relatively large, the scale of the model will be relatively large and the model performance will be relatively better. Of course, more time and resources will be consumed during inference or training; if the model parameters are relatively small, the scale of the model will be relatively small. When the performance meets the requirements, the model is more lightweight and consumes relatively less time and resources during inference or training.

[0044] Considering the limited computing resources of mobile terminals, the above method provided by the embodiments of the present application can be applied to, for example Figure 1 the application scenarios shown, but not limited thereto. Figure 1 is a schematic diagram of an application scenario of a method for generating feature data of an object according to an embodiment of the present application. As shown in Figure 1In the application scenario shown, the deep learning model is deployed in server 10. Server 10 can be connected to one or more client devices 20 through a local area network connection, a wide area network connection, an Internet connection, or other types of data networks. Here, client devices 20 can include, but are not limited to: smartphones, tablets, laptops, palmtop computers, personal computers, smart home devices, in-vehicle devices, etc. An operation interface for obtaining multimodal information of a target object can be deployed on the graphical user interface of the client device. Client device 20 can interact with the user through the graphical user interface to implement the invocation of the deep learning model, thereby implementing the method provided in the embodiments of the present application.

[0045] In the embodiments of the present application, the system composed of the client device and the server can perform the following steps: corresponding operations can be performed in the operation interface on client device 20 to obtain multimodal information of the target object. The client device can obtain the multimodal information of the target object and send it to the server through the network. After receiving the multimodal information of the target object, the server can perform the following steps: Step S102, obtain the multimodal information of the target object; Step S104, based on the multimodal information of the target object, determine at least one candidate object, where the similarity between the data distribution state of the candidate object and the data distribution state of the target object exceeds a first threshold; Step S106, obtain the feature data of the candidate object, where the feature data of the candidate object is used to represent the performance result of the candidate object on the object recommendation platform; Step S108, based on the feature data of the candidate object, predict the feature data of the target object, where the feature data of the target object is used to represent the performance result of the target object on the object recommendation platform.

[0046] It should be noted that with the rapid development of high-performance computing units, in other application scenarios, the above method provided in the embodiments of the present application can also be applied to a model all-in-one machine. In an optional embodiment, multiple models are built into the model all-in-one machine, and one model can be selected for adjustment as needed. Thus, the high-performance computing unit built into the model all-in-one machine can directly call the adjusted model to execute the above method provided in the embodiments of the present application. In another optional embodiment, a trained model is built into the deep learning model all-in-one machine. Thus, the high-performance computing unit built into the model all-in-one machine can directly call the model to execute the above method provided in the embodiments of the present application.

[0047] Further, when it is necessary to train a model in a target scenario, the client can also upload its own dataset, which is sent by the client to the server, enabling the server to adjust the pre-trained model with this dataset to obtain a model in the target scenario and then deploy it to the production environment. To facilitate the adjustment of the model, the server can provide a complete set of adjustment tools, development frameworks, and processes, supporting multiple adjustment strategies, so that the adjusted model can better adapt to different domain applications and achieve high customization.

[0048] Under the above operating environment, the present application provides a method for generating feature data of an object as shown in Figure 2 the following. Figure 2 FIG. is a flowchart of a method for generating feature data of an object according to an embodiment of the present application. As shown in Figure 2 the following, the method may include the following steps:

[0049] Step S202, obtaining multi-modal information of a target object.

[0050] In the technical solution provided in step S202 of the present application, multi-modal information of a target object can be obtained. Among them, the target object can be an object for which the performance result is to be predicted on an object recommendation platform. Optionally, the object can be an entity object or a non-entity object. Among them, the entity object can be an entity item. For example, the entity item can be an entity commodity in a transaction scenario, and the non-entity object can be a non-entity commodity. For example, the non-entity commodity can be a virtual commodity such as a digital collectible. Optionally, the target object can also be referred to as a commodity to be evaluated (such as a commodity to be scored), a new product to be evaluated, a new product in the field, etc. The multi-modal information can include various heterogeneous modal information (such as text information, picture information, etc.). For example, the multi-modal information can include text information such as the title and product introduction of the commodity to be scored, and / or include picture information such as the main picture of the commodity to be scored.

[0051] For example, the target object can be a commodity newly launched by a merchant or a commodity with little user interaction history (user feedback). The multi-modal information of the target object covers information in various expression forms related to the target object, including but not limited to images, text descriptions, prices, brand information, etc. By collecting and analyzing the multi-modal information of the target object, the performance result (such as the growth performance) of the target object on the object recommendation platform can be predicted to determine how to effectively recommend the target object.

[0052] Optionally, multimodal information of the target object can be collected from different data sources. For example, image information of the target object can be obtained from product detail pages, advertising images, or images uploaded by users. Text information of the target object can be obtained from product titles, descriptions, reviews, and user inquiry information. When there is an audio introduction or video demonstration in the product information, audio information or video information of the target object can be obtained from the audio introduction or video demonstration. Structured information of the target object can be obtained from product prices, sales volumes, inventory, brands, and categories. It should be noted that the above are only examples, and there are no specific restrictions on the acquisition methods and expression forms of the multimodal information of the target object.

[0053] Step S204: Determine at least one candidate object based on the multimodal information of the target object.

[0054] In the technical solution provided in step S204 of the present application, the similarity between the data distribution state of the candidate object and the data distribution state of the target object exceeds the first threshold.

[0055] In this embodiment, after obtaining the multimodal information of the target object, at least one candidate object can be determined based on the obtained multimodal information of the target object. Among them, the candidate object can be an object with a high similarity to the target object screened from the object set. For example, the candidate object can be a sequence of similar products retrieved from the object set, and can also be called a new product sequence, which can be represented by docs item. The object set can be a new product set, and can also be called previously exposed new products, previous new products, which can be represented by docs.

[0056] Optionally, the first threshold can be a critical value pre-set according to the actual situation for measuring the similarity between the data distribution state of the candidate object and the data distribution state of the target object. Since the similarity between the data distribution state of the candidate object and the data distribution state of the target object exceeds the first threshold, the data distribution state of the candidate object is consistent with the data distribution state of the target object.

[0057] Optionally, after obtaining the multimodal information of the target object, a vector retrieval service using approximate nearest neighbor (proxima) retrieval can be used to screen at least one candidate object from the object set whose data distribution state is consistent with that of the target object.

[0058] For example, after obtaining the multimodal information of the product to be scored, multiple candidate objects similar to the product to be scored and the similarity between each candidate object and the product to be scored can be retrieved from the previous new products docs.

[0059] Step S206: Obtain the feature data of the candidate object.

[0060] In the technical solution provided in step S206 of the present application, the feature data of the candidate object is used to represent the performance result of the candidate object on the object recommendation platform.

[0061] In this embodiment, after determining at least one candidate object based on the multimodal information of the target object, the feature data of the candidate object can be obtained. Among them, the feature data of the candidate object can be used to represent the performance result of the candidate object on the object recommendation platform, and this performance result can be the growth performance of the candidate object on the object recommendation platform, such as exposure, transaction, etc.

[0062] Optionally, after determining at least one candidate object based on the multimodal information of the target object, the performance result of the candidate object on the object recommendation platform can be statistically analyzed to obtain the feature data of the candidate object.

[0063] For example, after retrieving multiple new product sequences similar to the product to be scored from previous new product docs, the key growth performances of the new product sequences 10 days and 15 days after entering the field can be statistically analyzed, such as exposure, transaction, etc., so as to obtain the feature data of the new product sequences.

[0064] Step S208, predicting the feature data of the target object based on the feature data of the candidate object.

[0065] In the technical solution provided in step S208 of the present application, the feature data of the target object is used to represent the performance result of the target object on the object recommendation platform.

[0066] In this embodiment, after obtaining the feature data of the candidate object, based on the obtained feature data of the candidate object, the feature data of the target object can be predicted. Among them, the feature data of the target object can be used to represent the performance result of the target object on the object recommendation platform, and this performance result can be the growth performance of the target object on the object recommendation platform, such as exposure, transaction, etc. The feature data of the target object can also be referred to as the new product similarity sequence feature.

[0067] Optionally, when determining at least one candidate object based on the multimodal information of the target object, the similarity between each candidate object and the target object can also be determined to obtain multiple similarities. Furthermore, a normalization function (softmax function) can be used to convert the above obtained multiple similarity values into a probability distribution to obtain weights.

[0068] Optionally, after obtaining the feature data of the candidate object, the obtained weights can be used to perform weighted summation on the feature data of the candidate object (such as exposure, transaction, etc.), and the obtained weighted summation result can be determined as the feature data of the target object, that is, the new product similarity sequence feature used to characterize the growth of the new product is obtained.

[0069] Through the above steps S202 to S208 of this application, multi-modal information of the target object is obtained; based on the multi-modal information of the target object, at least one candidate object is determined, wherein the similarity between the data distribution state of the candidate object and the data distribution state of the target object exceeds a first threshold; characteristic data of the candidate object is obtained, wherein the characteristic data of the candidate object is used to represent the performance result of the candidate object on the object recommendation platform; based on the characteristic data of the candidate object, the characteristic data of the target object is predicted, wherein the characteristic data of the target object is used to represent the performance result of the target object on the object recommendation platform. That is to say, in the embodiment of this application, based on the multi-modal information of the target object, a candidate object with a data distribution state consistent with that of the target object is determined, and then, through the characteristic data of the candidate object, the characteristic data for characterizing the performance result of the target object is predicted, avoiding the situation that due to the lack of behavioral data of the target object, the characteristic data cannot be effectively determined based on the behavioral data of the target object, achieving the purpose of accurately describing the performance result of the target object, and further achieving the technical effect of improving the data generation efficiency of the object, and solving the technical problem of low data generation efficiency of the object.

[0070] Next, the method for determining at least one candidate object based on the multi-modal information of the target object in this embodiment will be further introduced.

[0071] As an optional implementation manner, the target object is an object published on the object recommendation platform during the current period. Step S204, based on the multi-modal information of the target object, determining at least one candidate object includes: determining, from the object set, candidate objects whose similarity between the multi-modal information and the multi-modal information of the target object exceeds a second threshold, where the object set includes objects published on the object recommendation platform during the historical period corresponding to the current period, and the similarity between the data distribution states of the objects in the object set and the target object on the object recommendation platform exceeds the first threshold.

[0072] In this embodiment, the target object may be an object published on the object recommendation platform during the current period. After obtaining the multi-modal information of the target object, candidate objects whose similarity between the multi-modal information and the multi-modal information of the target object exceeds a second threshold can be determined from the object set. Among them, the object set may include objects published on the object recommendation platform during the historical period corresponding to the current period, and the similarity between the data distribution states of the object and the target object on the object recommendation platform exceeds the first threshold, that is, the data distribution states of the target object and the objects in the object set are consistent.

[0073] Optionally, the second threshold may be a critical value set in advance according to the actual situation for measuring the similarity between the multimodal information of the candidate object and the multimodal information of the target object. For example, the similarity between the multimodal information of each candidate object and the multimodal information of the target object can be obtained. Further, the obtained multiple similarities can be sorted, and the similarity between the multimodal information of one of the candidate objects (for example, the k-th candidate object) and the multimodal information of the target object can be used as the second threshold. And there are k candidate objects in the object set whose similarity between the multimodal information and the multimodal information of the target object is greater than the second threshold. Thus, k candidate objects with high similarity between the multimodal information and the multimodal information of the target object can be selected from the object set. For example, when k is 50, the first 50 (top 50) new product sequences can be determined.

[0074] Optionally, after obtaining the multimodal information of the target object, the vector retrieval service of proxima retrieval can be used to screen candidate objects from the object set whose similarity between the multimodal information and the multimodal information of the target object exceeds the second threshold.

[0075] For example, after obtaining the multimodal information of the product to be scored, the vector retrieval service of proxima retrieval can be used to retrieve the top 50 new product sequences similar to the product to be scored from the previous new product docs, as well as the similarity between each candidate object and the product to be scored.

[0076] Since there is usually a lack of sufficient behavioral data in the initial stage of new product launch, traditional methods based on behavioral data to determine the characteristic data of new products are difficult to accurately evaluate and recommend new products. However, in this embodiment, the multimodal information of the new product is compared with the historical object set, and candidate objects with high similarity to the multimodal information of the new product can be determined, so as to better understand and predict the preferences of users for new products and provide more personalized and high-quality recommendation results.

[0077] The method for determining candidate objects from the object set whose similarity between the multimodal information and the multimodal information of the target object exceeds the second threshold in the above embodiment will be further introduced below.

[0078] As an optional implementation manner, determining candidate objects from the object set whose similarity between the multimodal information and the multimodal information of the target object exceeds the second threshold includes: converting the multimodal information of the target object into a multimodal feature vector of the target object; retrieving from the object set objects whose similarity between the multimodal feature vector and the multimodal feature vector of the target object exceeds the second threshold, where the multimodal feature vectors of the objects in the object set are obtained by converting the multimodal information of the objects in the object set; and determining the retrieved objects as candidate objects.

[0079] In this embodiment, after obtaining the multi-modal information of the target object, the obtained multi-modal information of the target object can be converted into a multi-modal feature vector of the target object. Further using the vector retrieval service, objects with a similarity exceeding the second threshold between the multi-modal feature vector and the multi-modal feature vector of the target object can be retrieved from the object set, and then the retrieved objects are determined as candidate objects. Among them, the multi-modal feature vector can also be referred to as a multi-modal representation.

[0080] Optionally, after obtaining the multi-modal information of the target object, the multi-modal information of the target object can be converted into a multi-modal representation of the target object. The multi-modal information of each object in the object set can also be converted into a multi-modal representation of each object. Further using the vector retrieval service of proxima, the topk candidate objects similar to the target object can be retrieved from the object set, and the similarity between the multi-modal representation of the candidate object and the multi-modal representation of the target object exceeds the second threshold.

[0081] For example, for the target object, after obtaining the multi-modal information of the target object, such as text information and picture information of the main product picture, title, product introduction, etc., the multi-modal information of the target object is converted into a multi-modal representation of the target object. For the object set, the multi-modal information of each object is converted into a multi-modal representation of each object. Further using the vector retrieval service of proxima, the top 50 candidate objects (i.e., the new product sequence) similar to the target object can be retrieved from the object set, and the similarity between the multi-modal representation of the candidate object and the multi-modal representation of the target object exceeds the second threshold.

[0082] This embodiment converts the multi-modal information of the target object into a multi-modal representation, and performs similarity analysis based on the multi-modal representation of the target object. The determined candidate objects do not rely solely on traditional single attributes or user behavior data. Even if a new product is just launched and there is not enough historical performance data, similar candidate objects can be retrieved through the multi-modal representation of the target object, thereby improving the retrieval efficiency and the accuracy of new product recommendations.

[0083] Next, a further introduction is made to the method of converting the multi-modal information of the target object into a multi-modal feature vector of the target object in this embodiment.

[0084] As an alternative implementation, converting the multi-modal information of the target object into a multi-modal feature vector of the target object includes: inputting the multi-modal information of the target object into a service inference model, and using the service inference model to infer the multi-modal information of the target object to obtain a multi-modal feature vector of the target object, where the service inference model is trained using multi-modal information samples.

[0085] In this embodiment, after obtaining the multimodal information of the target object, the obtained multimodal information of the target object can be input into the service inference model, and the service inference model is used to infer the multimodal information of the target object to obtain the multimodal feature vector of the target object. Among them, the service inference model can be trained using multimodal information samples and can be used to infer the multimodal information of the target object to obtain the multimodal feature vector.

[0086] For example, after obtaining the multimodal information of the target object, such as the main product image, title, product introduction, etc., which contains text information and image information, the service inference model can be used to infer the obtained multimodal information of the target object to obtain the multimodal representation of the target object.

[0087] In this embodiment, the service inference model is used to convert the multimodal information of the target object into a multimodal feature vector. Since the service inference model can simultaneously process different modalities of information, such as text information and image information, and convert the multimodal information of the target object into a unified multimodal feature vector, it is possible to comprehensively understand the characteristics of the target object, rather than just analyzing it based on a single modality. The fused multimodal feature vector contains the multi-dimensional representation of the target object, so that the semantic and visual features of the target object can be captured more accurately.

[0088] Next, a further introduction is made to the method of retrieving, from the object set, an object whose similarity between the multimodal feature vector and the multimodal feature vector of the target object exceeds the second threshold in this embodiment.

[0089] As an optional implementation manner, retrieving, from the object set, an object whose similarity between the multimodal feature vector and the multimodal feature vector of the target object exceeds the second threshold includes: calling a vector retrieval engine to retrieve, from the object set, an object whose similarity between the multimodal feature vector and the multimodal feature vector of the target object exceeds the second threshold, where the quality index of the multimodal feature vector of the candidate object meets the index threshold.

[0090] In this embodiment, after converting the multimodal information of the target object into the multimodal feature vector of the target object, a vector retrieval engine can be called to retrieve, from the object set, an object whose similarity between the multimodal feature vector and the multimodal feature vector of the target object exceeds the second threshold, and the retrieved object is determined as the candidate object. Among them, the vector retrieval engine can be an engine that performs similarity retrieval on the multimodal feature vector of the target object and the multimodal feature vectors of each object in the object set. For example, it can be a proxima retrieval engine or a similar product retrieval module.

[0091] Optionally, the quality index of the multi-modal feature vector of the candidate object meets the index threshold. The index threshold can be a critical value preset according to the actual situation for measuring the quality index of the multi-modal feature vector of the candidate object, that is, the multi-modal representation of the candidate object retrieved from the object set by invoking the vector retrieval engine is a high-quality multi-modal representation.

[0092] Optionally, after using the service inference model to infer the multi-modal information of the target object and obtaining the multi-modal feature vector of the target object, a vector retrieval engine can be invoked, that is, the vector retrieval service of proxima, to screen out candidate objects from the object set whose similarity between the multi-modal information and the multi-modal information of the target object exceeds the second threshold. For example, the top 50 similar candidate objects can be screened out, and the multi-modal representations of the screened candidate objects are high-quality commodity multi-modal representations.

[0093] This embodiment retrieves the multi-modal feature vectors of similar objects through a vector retrieval engine and controls the quality of the retrieved multi-modal feature vectors. Quality control can avoid inaccurate recommendation results caused by multi-modal feature vector errors, thereby providing higher-quality recommendation results.

[0094] The method for obtaining the feature data of the candidate object in the above embodiment will be further introduced below.

[0095] As an optional implementation manner, step S206, obtaining the feature data of the candidate object, includes: determining, in the historical period, the target historical period when the candidate object is published to the object recommendation platform; obtaining the feature sequence of the candidate object in the target historical period, where the feature sequence includes feature data in different data representation dimensions, and the feature sequence is used to represent the performance result of the candidate object on the object recommendation platform in the target historical period.

[0096] In this embodiment, after determining at least one candidate object based on the multi-modal information of the target object, the target historical period when the candidate object is published to the object recommendation platform can be determined in the historical period. Further, the feature sequence of the candidate object in the target historical period can be obtained. Among them, the historical period can be the duration when the candidate object is published to the object recommendation platform. For example, the historical period can be 5 days, 10 days, 15 days, 20 days, etc. The target historical period can be the duration selected from the historical period. For example, the target historical period can be 10 days, 15 days. It should be noted that the above is only an example, and there is no specific limitation on the duration of the historical period and the target historical period.

[0097] Optionally, the feature sequence may include feature data under different data performance dimensions. For example, the data performance dimensions may be dimensions such as exposure and transaction. The feature sequence can be used to represent the performance results of a candidate object on the object recommendation platform during a target historical period.

[0098] Optionally, after determining the candidate object, the target historical period when the candidate object is published on the object recommendation platform can be determined. Further, the feature data of the candidate object under different data performance dimensions (such as exposure, transaction, etc.) during the target historical period can be statistically analyzed to obtain the feature sequence of the candidate object during the target historical period.

[0099] For example, the target historical period is 10 days or 15 days. After using the vector retrieval service of Proxima to screen out the top 50 similar candidate objects from the object set, the feature sequences of the candidate objects during the target historical periods of 10 days and 15 days can be obtained. The key growth performances of the candidate objects in the 10 days and 15 days after entering the field, such as exposure and transaction, can be statistically analyzed to obtain the feature sequences of the candidate objects during the target historical periods of 10 days and 15 days.

[0100] In this embodiment, by determining the target historical period when the candidate object is published on the object recommendation platform and then analyzing the feature sequence of the candidate object during the target historical period, the feature data of the target object under similar conditions can be predicted, and thus more accurate personalized recommendations can be provided.

[0101] Next, the method for predicting the feature data of the target object based on the feature data of the candidate object in this embodiment will be further introduced.

[0102] As an optional implementation manner, step S208, predicting the feature data of the target object based on the feature data of the candidate object, includes: predicting the feature sequence of the target object based on the feature sequence of the candidate object and the similarity corresponding to the candidate object, where the similarity corresponding to the candidate object is used to represent the similarity exceeding the second threshold between the multimodal information of the candidate object and the multimodal information of the target object, and the feature sequence of the target object is used to represent the performance result of the object on the object recommendation platform during the future period corresponding to the current period.

[0103] In this embodiment, after obtaining the feature sequence of the candidate object in the target historical period, the feature sequence of the target object can be predicted based on the obtained feature sequence of the candidate object and the similarity corresponding to the candidate object. Among them, the similarity corresponding to the candidate object can be used to represent the similarity exceeding the second threshold between the multi-modal information of the candidate object and the multi-modal information of the target object, and this similarity can be represented by sim_score. The feature sequence of the target object can be used to represent the performance result of the object on the object recommendation platform in the future period corresponding to the current period.

[0104] Optionally, while screening out candidate objects similar to the target object from the object set using the vector retrieval service of proxima, the similarity between each candidate object and the target object can also be obtained. After obtaining the feature sequence of the candidate object in the target historical period, the feature sequence of the target object can be predicted based on the obtained feature sequence of the candidate object and the similarity between each candidate object and the target object.

[0105] Optionally, after obtaining the feature sequence of the candidate object in the target historical period, the similarity between each candidate object and the target object can be normalized to obtain weights, and then the weights are used to perform weighted summation on the feature sequence of the candidate object in the target historical period to obtain a new product similarity sequence feature for characterizing new product growth.

[0106] By analyzing the performance characteristics (such as exposure, transactions, etc.) of candidate objects similar to the target object in multi-modal information in the target historical period, this embodiment can predict the growth path of the target object in the future period, thereby greatly improving the efficiency and accuracy of new product management and recommendation.

[0107] Next, a further introduction is made to the method of predicting the feature sequence of the target object based on the feature sequence of the candidate object and the similarity corresponding to the candidate object in this embodiment.

[0108] As an alternative implementation manner, predicting the feature sequence of the target object based on the feature sequence of the candidate object and the similarity corresponding to the candidate object includes: respectively normalizing the similarities corresponding to multiple candidate objects to obtain weights corresponding to the multiple candidate objects, where the weights are used to represent the importance degree of the feature sequence of the corresponding candidate object to the feature sequence of the target object; using the weights corresponding to the multiple candidate objects to merge the feature sequences of the multiple candidate objects to obtain the feature sequence of the target object.

[0109] In this embodiment, the similarities corresponding to multiple candidate objects can be normalized respectively to obtain the weights corresponding to the multiple candidate objects. Then, after obtaining the feature sequences of the candidate objects in the target historical period, the feature sequences of the multiple candidate objects can be merged by using the weights corresponding to the multiple candidate objects to obtain the feature sequence of the target object.

[0110] Optionally, after obtaining the similarities between each candidate object and the target object, a normalization function (softmax function) can be used to normalize the multiple similarities obtained above to obtain the weights corresponding to the multiple candidate objects. Further, the weighted sum of the feature sequences of the multiple candidate objects is calculated by using the obtained weights corresponding to the multiple candidate objects to obtain the feature sequence of the target object.

[0111] For example, after obtaining the similarity sim_score between each candidate object and the target object, the softmax function can be used to convert the multiple similarities obtained above into a probability distribution, ensuring that the values of the output results are between 0 and 1 and the sum of the output results is 1, so as to obtain the weight weight, that is, weight = softmax(sim_score). Finally, the obtained weight is used to calculate the weighted sum of the feature sequences of the candidate objects in different data performance dimensions (such as exposure, transaction, etc.) to obtain the new product similarity sequence feature for characterizing the growth of the new product.

[0112] In this embodiment, the weights are obtained by normalizing multiple similarities, and then the weighted sum of the first-guessed new product growth performance data of the new product sequence is calculated by using the obtained weights to obtain the new product similarity sequence feature, so as to accurately predict the growth trajectory and performance results of the new product on the recommendation platform.

[0113] Next, a further introduction is made to the method of merging the feature sequences of multiple candidate objects by using the weights corresponding to the multiple candidate objects to obtain the feature sequence of the target object in this embodiment.

[0114] As an alternative implementation, merging the feature sequences of multiple candidate objects by using the weights corresponding to the multiple candidate objects to obtain the feature sequence of the target object includes: calculating the weighted sum of the feature sequences of the multiple candidate objects by using the weights corresponding to the multiple candidate objects to obtain a summation result; splicing the summation result with the feature sequences of at least one auxiliary object to obtain the feature sequence of the target object, where the auxiliary object is an object other than the object set published on the object recommendation platform in the historical period.

[0115] In this embodiment, after normalizing the similarities corresponding to multiple candidate objects respectively to obtain the weights corresponding to the multiple candidate objects, the weighted sum of the feature sequences of the multiple candidate objects can be calculated using the weights corresponding to the multiple candidate objects to obtain a summation result. Further, the summation result is concatenated with the feature sequences of at least one auxiliary object to obtain the feature sequence of the target object. Herein, the auxiliary object can be an object other than the object set published on the object recommendation platform during the historical period. The feature sequence of the auxiliary object can be used to represent the performance result of the auxiliary object on the object recommendation platform during the historical period. For example, the feature sequence of the auxiliary object can be the static statistical attributes, generalization features, statistical features, multi-modal features, similarity sequences of the first-guessed old products, etc. of the auxiliary object. This is only an example here and does not specifically limit the feature sequence of the auxiliary object.

[0116] For example, after normalizing the obtained multiple similarities using the softmax function to obtain the weight weight, the weighted sum of the feature sequences of the candidate object in different data performance dimensions (such as exposure, transaction, etc.) can be calculated using the obtained weight to obtain a summation result. Further, the summation result is concatenated with other product features (such as static statistical attributes, generalization features, statistical features, multi-modal features, similarity sequences of the first-guessed old products, etc.) to obtain the similarity sequence feature of the new product for characterizing the growth of the new product.

[0117] The weights obtained in this embodiment can reflect the importance degree of each candidate object for predicting the feature sequence of the target object. By calculating the weighted sum of the feature sequences of the candidate objects using the weights, the comprehensiveness and accuracy of the predicted feature sequence of the target object can be ensured.

[0118] Next, the method of concatenating the summation result with the feature sequences of at least one auxiliary object to obtain the feature sequence of the target object in this embodiment will be further introduced.

[0119] As an optional implementation manner, concatenating the summation result with the feature sequences of at least one auxiliary object to obtain the feature sequence of the target object includes: inputting the summation result and the feature sequences of the auxiliary object into a feature prediction model, and using the feature prediction model to concatenate the summation result and the feature sequences of the auxiliary object to obtain the feature sequence of the target object, wherein the feature prediction model is trained using the feature sequence samples of object samples.

[0120] In this embodiment, after obtaining the summation result by performing weighted summation on the feature sequences of multiple candidate objects using the weights corresponding to the multiple candidate objects, the obtained summation result and the feature sequence of the auxiliary object can be input into the feature prediction model, and the feature prediction model is used to splice the summation result and the feature sequence of the auxiliary object to obtain the feature sequence of the target object. Among them, the feature prediction model can be trained using the feature sequence samples of object samples, and the feature prediction model can also be referred to as a new product potential estimation model.

[0121] Optionally, after obtaining the summation result by performing weighted summation on the feature sequences of the candidate objects in different data performance dimensions using the obtained weights, the summation result and other product features (such as static statistical attributes, generalization features, statistical features, multi-modal features, similar sequences of the first guessed old products, etc.) can be input into the new product potential estimation model, and the new product potential estimation model is used to splice the summation result and other product features to obtain the new product similar sequence feature for characterizing the growth of the new product.

[0122] In this embodiment, by splicing the summation result of performing weighted summation on the feature sequences of multiple candidate objects and other product features, a new product similar sequence feature containing multi-dimensional information is generated. This new product similar sequence feature not only covers the static information of the new product (such as price, category, brand, etc.), but also includes the dynamic prediction of the growth potential and market performance of the new product, thereby more accurately evaluating the comprehensive value and market potential of the new product.

[0123] In the embodiment of the present application, multi-modal information of the target object is obtained; based on the multi-modal information of the target object, at least one candidate object is determined, where the similarity between the data distribution state of the candidate object and the data distribution state of the target object exceeds a first threshold; feature data of the candidate object is obtained, where the feature data of the candidate object is used to represent the performance result of the candidate object on the object recommendation platform; based on the feature data of the candidate object, feature data of the target object is predicted, where the feature data of the target object is used to represent the performance result of the target object on the object recommendation platform. That is to say, in the embodiment of the present application, based on the multi-modal information of the target object, a candidate object with a data distribution state consistent with that of the target object is determined, and then, through the feature data of the candidate object, the feature data for characterizing the performance result of the target object is predicted, avoiding the situation that due to the lack of behavioral data of the target object, the feature data cannot be effectively determined based on the behavioral data of the target object, achieving the purpose of accurately describing the performance result of the target object, and further realizing the technical effect of improving the data generation efficiency of the object and solving the technical problem of low data generation efficiency of the object.

[0124] The embodiment of the present application also provides a method for generating feature data of an object. Figure 3It is a flowchart of a method for generating feature data of another object according to an embodiment of the present application. As Figure 3 shown, the method may include the following steps:

[0125] Step S302, obtain multimodal information of a target transaction object.

[0126] In the technical solution provided in step S302 of the present application above, multimodal information of a target transaction object can be obtained. Among them, the target transaction object may be a selected transaction object for predicting the performance result on an e-commerce platform. For example, the transaction object may be a traded commodity.

[0127] For example, the target transaction object may be a traded commodity that a merchant has just launched or has little user interaction history (user feedback). The multimodal information of the target transaction object covers information in various expression forms related to the target transaction object, including but not limited to images, text descriptions, prices, brand information, etc. By collecting and analyzing the multimodal information of the target transaction object, the performance result (such as growth performance) of the target transaction object on the e-commerce platform can be predicted to determine how to effectively recommend the target transaction object.

[0128] Optionally, multimodal information of the target transaction object can be collected from different data sources. For example, image information of the target transaction object can be obtained from a product detail page, an advertisement picture, or a picture uploaded by a user. Text information of the target transaction object can be obtained from a product title, description, comment, or user inquiry. When there is an audio introduction or video demonstration in the product information, audio information or video information of the target transaction object can be obtained from the audio introduction or video demonstration. Structured information of the target transaction object can be obtained from a product price, sales volume, inventory, brand, or category. It should be noted that the above is only an example, and there is no specific limitation on the acquisition method and expression form of the multimodal information of the target transaction object.

[0129] Step S304, determine at least one candidate transaction object based on the multimodal information of the target transaction object.

[0130] In the technical solution provided in step S304 of the present application above, the similarity between the data distribution state of the candidate transaction object and the data distribution state of the target transaction object exceeds a first threshold.

[0131] In this embodiment, after obtaining the multimodal information of the target transaction object, at least one candidate transaction object can be determined based on the obtained multimodal information of the target transaction object. Among them, the candidate transaction object may be a transaction object with a high similarity to the target transaction object selected from a set of transaction objects. For example, the candidate transaction object may be a sequence of similar traded commodities retrieved from a set of transaction objects.

[0132] Optionally, the first threshold value can be a critical value preset according to the actual situation for measuring the similarity between the data distribution state of the candidate trading object and the data distribution state of the target trading object. Since the similarity between the data distribution state of the candidate trading object and the data distribution state of the target trading object exceeds the first threshold value, the data distribution state of the candidate trading object is consistent with the data distribution state of the target trading object.

[0133] Optionally, after obtaining the multimodal information of the target trading object, the vector retrieval service of proxima retrieval can be used to screen at least one candidate trading object with a data distribution state consistent with that of the target trading object from the trading object set.

[0134] Step S306: Obtain the feature data of the candidate trading object.

[0135] In the technical solution provided in step S306 of the present application above, the feature data of the candidate trading object is used to represent the performance result of the candidate trading object on the e-commerce platform.

[0136] In this embodiment, after determining at least one candidate trading object based on the multimodal information of the target trading object, the feature data of the candidate trading object can be obtained. Among them, the feature data of the candidate trading object can be used to represent the performance result of the candidate trading object on the e-commerce platform, and this performance result can be the growth performance of the candidate trading object on the e-commerce platform, such as exposure, transaction, etc.

[0137] Optionally, after determining at least one candidate trading object based on the multimodal information of the target trading object, the performance result of the candidate trading object on the e-commerce platform can be statistically analyzed to obtain the feature data of the candidate trading object.

[0138] Step S308: Predict the feature data of the target trading object based on the feature data of the candidate trading object.

[0139] In the technical solution provided in step S308 of the present application above, the feature data of the target trading object is used to represent the performance result of the target trading object on the e-commerce platform.

[0140] In this embodiment, after obtaining the feature data of the candidate trading object, based on the obtained feature data of the candidate trading object, the feature data of the target trading object can be predicted. Among them, the feature data of the target trading object can be used to represent the performance result of the target trading object on the e-commerce platform, and this performance result can be the growth performance of the target trading object on the e-commerce platform, such as exposure, transaction, etc.

[0141] Optionally, while determining at least one candidate trading object based on the multimodal information of the target trading object, the similarity between each candidate trading object and the target trading object can also be determined to obtain multiple similarities. Furthermore, the softmax function can be used to convert the above-obtained multiple similarity values into a probability distribution to obtain the weight weight.

[0142] Optionally, after obtaining the feature data of the candidate trading object, the obtained weight can be used to perform weighted summation on the feature data of the candidate trading object (such as exposure, transaction, etc.) to obtain the feature data of the target trading object.

[0143] Through the above steps S302 to S308 of this application, the multimodal information of the target trading object is obtained; based on the multimodal information of the target trading object, at least one candidate trading object is determined, where the similarity between the data distribution state of the candidate trading object and the data distribution state of the target trading object exceeds a first threshold; the feature data of the candidate trading object is obtained, where the feature data of the candidate trading object is used to represent the performance result of the candidate trading object on the e-commerce platform; based on the feature data of the candidate trading object, the feature data of the target trading object is predicted, where the feature data of the target trading object is used to represent the performance result of the target trading object on the e-commerce platform. That is to say, the embodiment of this application determines a candidate trading object whose data distribution state is consistent with that of the target trading object based on the multimodal information of the target trading object, and then predicts the feature data used to characterize the performance result of the target trading object through the feature data of the candidate trading object, avoiding the situation that due to the lack of behavioral data of the target trading object, the feature data cannot be effectively determined based on the behavioral data of the target trading object, achieving the purpose of accurately describing the performance result of the target trading object, and further realizing the technical effect of improving the data generation efficiency of the object, and solving the technical problem of low data generation efficiency of the object.

[0144] The above method of this embodiment will be further introduced below.

[0145] As an optional implementation manner, the method further includes: based on the feature data of the target trading object, displaying the target trading object in the display area on the e-commerce platform, where the display priority of the display area is higher than that of the areas other than the display area on the e-commerce platform.

[0146] In this embodiment, after predicting the characteristic data of the target trading object based on the characteristic data of the candidate trading object, based on the predicted characteristic data of the target trading object, the target trading object can be displayed in the display area on the e-commerce platform. Among them, the display priority of the display area is higher than that of the areas other than the display area on the e-commerce platform. For example, the display area can be the home page of the e-commerce platform.

[0147] For example, after predicting the characteristic data of the target trading object based on the characteristic data of the candidate trading object, based on the predicted characteristic data of the target trading object, the target trading object can be displayed on the home page of the e-commerce platform. That is, the target trading object is recommended on the home page.

[0148] Based on the predicted characteristic data of the target trading object, this embodiment displays the target trading object on the home page of the e-commerce platform, which can adjust the home page layout and recommendation strategy of the e-commerce platform to meet the user needs.

[0149] Next, a further introduction is made to the method of displaying the target trading object in the display area on the e-commerce platform based on the characteristic data of the target trading object in this embodiment.

[0150] As an alternative implementation, displaying the target trading object in the display area on the e-commerce platform based on the characteristic data of the target trading object includes: determining the value attribute corresponding to the characteristic data of the target trading object; in response to the value attribute meeting the value attribute threshold, displaying the target trading object in the display area on the e-commerce platform.

[0151] In this embodiment, after predicting the characteristic data of the target trading object based on the characteristic data of the candidate trading object, the value attribute corresponding to the characteristic data of the target trading object can be determined. When the determined value attribute meets the value attribute threshold, the target trading object can be displayed in the display area on the e-commerce platform. Among them, the value attribute corresponding to the characteristic data of the target trading object can be used to represent the value or potential of the target trading object. The value attribute threshold can be a critical value preset according to the actual situation for measuring the value attribute corresponding to the characteristic data of the target trading object.

[0152] Optionally, after predicting the characteristic data of the target trading object based on the characteristic data of the candidate trading object, the value attribute corresponding to the characteristic data of the target trading object can be determined. When the determined value attribute meets the value attribute threshold, it indicates that the target trading object is a target trading object with high potential or high value, and then the target trading object with high potential or high value is displayed on the home page of the e-commerce platform.

[0153] In this embodiment, by displaying target trading objects whose value attributes corresponding to feature data meet the value attribute threshold on the home page of the e-commerce platform, users can come into contact with these target trading objects with high potential or high value in the first place, reducing the search and decision-making time of users on the platform, thereby improving the shopping efficiency and satisfaction of users.

[0154] In the embodiment of the present application, based on the multimodal information of the target trading object, candidate trading objects consistent with the data distribution state of the target trading object are determined, and then, through the feature data of the candidate trading objects, the feature data used to characterize the performance result of the target trading object is predicted, avoiding the situation that due to the lack of behavioral data of the target trading object, the feature data cannot be effectively determined based on the behavioral data of the target trading object, achieving the purpose of accurately describing the performance result of the target trading object, and further achieving the technical effect of improving the data generation efficiency of the object, and solving the technical problem of low data generation efficiency of the object.

[0155] The embodiment of the present application also provides a method for generating feature data of an object. Figure 4 It is a flowchart of another method for generating feature data of an object according to the embodiment of the present application, as Figure 4 shown, the method may include the following steps:

[0156] Step S402, in response to an input operation on the operation interface, display the multimodal information of the target object on the operation interface.

[0157] In the technical solution provided in step S402 of the present application above, in response to an input operation on the operation interface, the multimodal information of the target object can be displayed on the operation interface. Among them, the operation interface may be an operation interface displayed on a client device. For example, it may be an interface for interaction between a user and a computer or software system, and may be a graphical user interface, a command-line interface, etc. This is only for illustration and does not specifically limit the form of the operation interface.

[0158] In this embodiment, the input instruction may be an instruction for controlling the operation interface to display the multimodal information of the target object. For example, it may be an instruction input by the user on the operation interface. The user can generate the input instruction by inputting text through the keyboard, or clicking a button or option on the operation interface through the mouse, or dragging the multimodal information of the target object to a specified area.

[0159] In an alternative embodiment, the user can open the operation interface deployed on the client device and input the multimodal information of the target object on the operation interface to generate an input instruction. According to the generated input instruction, the corresponding multimodal information of the target object can be loaded and the multimodal information of the target object can be displayed on the operation interface.

[0160] Step S404: In response to a feature prediction operation on the operation interface, display the feature data of the target object on the operation interface.

[0161] In the technical solution provided in step S404 of the present application above, the feature data of the target object is predicted based on the feature data of at least one candidate object, and is used to represent the performance result of the target object on the object recommendation platform. The feature data of the candidate object is used to represent the performance result of the candidate object on the object recommendation platform. The candidate object is determined based on the multimodal information of the target object, and the similarity between the data distribution state of the candidate object and the data distribution state of the target object exceeds a first threshold.

[0162] In this embodiment, in response to a feature prediction operation on the operation interface, the feature data of the target object can be displayed on the operation interface. Among them, the feature prediction operation can be an operation of predicting the features of the multimodal information of the target object displayed on the operation interface. The user can perform the feature prediction operation by clicking on the multimodal information of the target object displayed on the operation interface. This is only an example here and does not specifically limit the feature prediction operation.

[0163] Optionally, after displaying the multimodal information of the target object on the operation interface, in response to a feature prediction operation on the operation interface, the vector retrieval service of proxima can be used to screen at least one candidate object from the object set that is consistent with the data distribution state of the target object. After determining at least one candidate object based on the multimodal information of the target object, the performance results of the candidate objects on the object recommendation platform can be statistically analyzed to obtain the feature data of the candidate objects.

[0164] Optionally, when determining at least one candidate object based on the multimodal information of the target object, the similarity between each candidate object and the target object can also be determined to obtain multiple similarities. Furthermore, the softmax function can be used to convert the above-obtained multiple similarity values into a probability distribution to obtain the weight weight. After obtaining the feature data of the candidate objects, the obtained weight can be used to perform a weighted sum of the feature data of the candidate objects (such as exposure, transactions, etc.) to obtain the feature data of the target object, that is, to obtain the new product similarity sequence feature used to characterize the growth of new products.

[0165] Through the above steps S402 to S404 of this application, in response to an input operation on the operation interface, multi-modal information of the target object is displayed on the operation interface; in response to a feature prediction operation on the operation interface, feature data of the target object is displayed on the operation interface, where the feature data of the target object is predicted based on the feature data of at least one candidate object and is used to represent the performance result of the target object on the object recommendation platform, the feature data of the candidate object is used to represent the performance result of the candidate object on the object recommendation platform, the candidate object is determined based on the multi-modal information of the target object, and the similarity between the data distribution state of the candidate object and the data distribution state of the target object exceeds a first threshold. That is to say, in the embodiment of this application, in response to a feature prediction operation on the operation interface, feature data of the target object is displayed on the operation interface, the feature data of the target object is predicted based on the feature data of at least one candidate object, the candidate object is determined based on the multi-modal information of the target object and is consistent with the data distribution state of the target object, avoiding the situation that due to the lack of behavior data of the target object, the feature data cannot be effectively determined based on the behavior data of the target object, achieving the purpose of accurately describing the performance result of the target object, and further realizing the technical effect of improving the data generation efficiency of the object and solving the technical problem of low data generation efficiency of the object.

[0166] The embodiment of this application also provides a method for generating feature data of an object. Figure 5 It is a flowchart of another method for generating feature data of an object according to the embodiment of this application, as Figure 5 shown. This method may include the following steps:

[0167] Step S502, obtain the multi-modal information of the target object by calling the first interface.

[0168] In the technical solution provided in step S502 of this application above, the first interface includes a first parameter, and the parameter value of the first parameter includes the multi-modal information of the target object.

[0169] In this embodiment, by calling the first interface, the multi-modal information of the target object can be obtained. Among them, the first interface may include a first parameter, and the parameter value of the first parameter may include the multi-modal information of the target object.

[0170] For example, the target object may be a product just launched by a merchant or with little user interaction history (user feedback). The multi-modal information of the target object covers information in various expression forms related to the target object, including but not limited to images, text descriptions, prices, brand information, etc. By collecting and analyzing the multi-modal information of the target object, the performance result (such as growth performance) of the target object on the object recommendation platform can be predicted to determine how to effectively recommend the target object.

[0171] Optionally, multimodal information of the target object can be collected from different data sources. For example, image information of the target object can be obtained from product detail pages, advertising pictures, or pictures uploaded by users. Text information of the target object can be obtained from product titles, descriptions, reviews, and user inquiries. When there is an audio introduction or video demonstration in the product information, audio information or video information of the target object can be obtained from the audio introduction or video demonstration. Structured information of the target object can be obtained from product prices, sales volumes, inventory, brands, and categories. It should be noted that the above are only examples, and there are no specific restrictions on the acquisition methods and expression forms of the multimodal information of the target object.

[0172] Step S504: Determine at least one candidate object based on the multimodal information of the target object.

[0173] In the technical solution provided in step S504 of the present application, the similarity between the data distribution state of the candidate object and the data distribution state of the target object exceeds a first threshold.

[0174] In this embodiment, after obtaining the multimodal information of the target object by calling the first interface, at least one candidate object can be determined based on the obtained multimodal information of the target object.

[0175] Optionally, after obtaining the multimodal information of the target object, a vector retrieval service of proxima retrieval can be used to screen at least one candidate object from the object set that is consistent with the data distribution state of the target object.

[0176] For example, after obtaining the multimodal information of the product to be scored, multiple candidate objects similar to the product to be scored and the similarity between each candidate object and the product to be scored can be retrieved from previous new product docs.

[0177] Step S506: Obtain the feature data of the candidate object.

[0178] In the technical solution provided in step S506 of the present application, the feature data of the candidate object is used to represent the performance result of the candidate object on the object recommendation platform.

[0179] In this embodiment, after determining at least one candidate object based on the multimodal information of the target object, the feature data of the candidate object can be obtained.

[0180] Optionally, after determining at least one candidate object based on the multimodal information of the target object, the performance result of the candidate object on the object recommendation platform can be statistically analyzed to obtain the feature data of the candidate object.

[0181] For example, after retrieving multiple new product sequences similar to the product to be scored from previous new product docs, the key growth performance of the new product sequences in the 10th and 15th days after entering the field can be counted, such as exposure, transactions, etc., so as to obtain the characteristic data of the new product sequences.

[0182] Step S508, based on the characteristic data of the candidate object, predict the characteristic data of the target object.

[0183] In the technical solution provided in step S508 of the present application above, the characteristic data of the target object is used to represent the performance result of the target object on the object recommendation platform.

[0184] In this embodiment, after obtaining the characteristic data of the candidate object, based on the obtained characteristic data of the candidate object, the characteristic data of the target object can be predicted.

[0185] Optionally, when determining at least one candidate object based on the multi-modal information of the target object, the similarity between each candidate object and the target object can also be determined to obtain multiple similarities. Furthermore, the softmax function can be used to convert the above-obtained multiple similarity values into a probability distribution to obtain the weight weight.

[0186] Optionally, after obtaining the characteristic data of the candidate object, the obtained weight can be used to perform a weighted sum of the characteristic data of the candidate object (such as exposure, transactions, etc.) to obtain the characteristic data of the target object, that is, to obtain the characteristic of the new product similarity sequence for characterizing the growth of the new product.

[0187] Step S510, output the characteristic data of the target object by calling the second interface.

[0188] In the technical solution provided in step S510 of the present application above, the second interface includes a second parameter, and the parameter value of the second parameter includes the characteristic data of the target object.

[0189] In this embodiment, after predicting the characteristic data of the target object based on the characteristic data of the candidate object, the characteristic data of the target object can be output by calling the second interface. Among them, the second interface may include a second parameter, and the parameter value of the second parameter may include the characteristic data of the target object.

[0190] Through the above steps S502 to S510 of this application, by invoking the first interface, multi-modal information of the target object is obtained. The first interface includes a first parameter, and the parameter value of the first parameter includes the multi-modal information of the target object. Based on the multi-modal information of the target object, at least one candidate object is determined, where the similarity between the data distribution state of the candidate object and the data distribution state of the target object exceeds a first threshold. Feature data of the candidate object is obtained, where the feature data of the candidate object is used to represent the performance result of the candidate object on the object recommendation platform. Based on the feature data of the candidate object, the feature data of the target object is predicted, where the feature data of the target object is used to represent the performance result of the target object on the object recommendation platform. The feature data of the target object is output by invoking the second interface. The second interface includes a second parameter, and the parameter value of the second parameter includes the feature data of the target object. That is to say, in the embodiment of this application, based on the multi-modal information of the target object, a candidate object with a data distribution state consistent with that of the target object is determined. Then, through the feature data of the candidate object, the feature data used to characterize the performance result of the target object is predicted, avoiding the situation that due to the lack of behavioral data of the target object, the feature data cannot be effectively determined based on the behavioral data of the target object, achieving the purpose of accurately describing the performance result of the target object, and further realizing the technical effect of improving the data generation efficiency of the object, and solving the technical problem of low data generation efficiency of the object.

[0191] The embodiment of this application also provides a system for generating feature data of an object. Figure 6 It is a schematic diagram of a system for generating feature data of an object according to an embodiment of this application, as Figure 6 shown. The system 600 for generating feature data of an object may include: a client 602 and a server 604.

[0192] The client 602 is used to transmit the multi-modal information of the target object.

[0193] The server 604 is used to determine at least one candidate object based on the multi-modal information of the target object, where the similarity between the data distribution state of the candidate object and the data distribution state of the target object exceeds a first threshold; obtain the feature data of the candidate object, where the feature data of the candidate object is used to represent the performance result of the candidate object on the object recommendation platform; predict the feature data of the target object based on the feature data of the candidate object, where the feature data of the target object is used to represent the performance result of the target object on the object recommendation platform; and return the feature data of the target object to the client.

[0194] In the feature data generation system 600 of the object, the multimodal information of the target object is transmitted through the client 602. Based on the multimodal information of the target object, the server 604 determines at least one candidate object, where the similarity between the data distribution state of the candidate object and the data distribution state of the target object exceeds a first threshold; obtains the feature data of the candidate object, where the feature data of the candidate object is used to represent the performance result of the candidate object on the object recommendation platform; predicts the feature data of the target object based on the feature data of the candidate object, where the feature data of the target object is used to represent the performance result of the target object on the object recommendation platform; and returns the feature data of the target object to the client. That is to say, the feature data generation system of the object determines a candidate object whose data distribution state is consistent with that of the target object based on the multimodal information of the target object, and then predicts the feature data used to characterize the performance result of the target object through the feature data of the candidate object, avoiding the situation that due to the lack of behavioral data of the target object, the feature data cannot be effectively determined based on the behavioral data of the target object, achieving the purpose of accurately describing the performance result of the target object, and further realizing the technical effect of improving the data generation efficiency of the object, and solving the technical problem of low data generation efficiency of the object.

[0195] The technical solutions of the embodiments of the present disclosure will be further introduced by way of preferred embodiments below.

[0196] Currently, when selecting high-quality products from a large number of new products, a dedicated product growth path can be used to provide certain support for these products. After the support, the products can grow faster and gain stronger competitiveness in the entire home page recommended product pool.

[0197] In an optional example, the weights of sequence features are usually obtained through model learning using the behavioral data in the system, and then the features of the products in the sequence are weighted and fused according to the weights. However, due to the lack of behavioral data of new products, this method cannot obtain effective sequence feature weights. In addition, the internal old products of the multimodal vector retrieval system are usually used to calculate the similarity to obtain the sequence feature weights, and finally the product features of the old products are weighted and fused to represent the new products. Although this method can solve the above-mentioned problem of insufficient behavioral data of new products, from the perspective of data distribution, there are significant differences between new products and old products themselves, and there is a problem that there are differences between using the recent features of old products to represent new products and the true performance of new products.

[0198] To solve the above problems, the present application proposes a new product characterization extraction method based on multi-modal similar sequences. By retrieving similar sequences based on multi-modal, this new product characterization extraction method in this field improves the algorithm's ability to depict the potential and growth rate of new products, thereby enhancing the product selection ability in the first-guess field. This method can be applied to the growth link of first-guess products to ensure the harmonious development of the ecosystem of the first-guess product pool.

[0199] Figure 7 It is a schematic diagram of generating new product similar sequence features according to an embodiment of the present application, as Figure 7 shown. For the new product to be evaluated (the product to be scored) and the previously exposed new products (the new product set), the multi-modal representations of the new product query and the previously exposed new product docs are obtained through the service inference model. That is, the service inference model is used to convert the multi-modal information of the product to be scored and the new product set into multi-modal representations respectively. For example, using text and picture information such as the main product image, title, and product introduction, the service inference model uses the multi-modal representation inference service to convert the product to be scored and the new product set into multi-modal features.

[0200] Use the vector retrieval service of proxima to retrieve the top k (for example, k = 50, that is, top 50) new product sequences and their similarities similar to the product query to be scored in the previously new product docs: query item: [(docs item1, sim_score1), (docs item2, sim_score2),..., (docs itemk, sim_scorek)].

[0201] Among them, query item can be used to represent the product sequence to be scored, docs item1 can be used to represent the first new product sequence, and sim_score1 can be used to represent the similarity between the product sequence to be scored and the first new product sequence; docs item2 can be used to represent the second new product sequence, and sim_score2 can be used to represent the similarity between the product sequence to be scored and the second new product sequence; docs itemk can be used to represent the k-th new product sequence, and sim_scorek can be used to represent the similarity between the product sequence to be scored and the k-th new product sequence.

[0202] Using proxima retrieval, from the multi-modal representations of the new product set, retrieve the top k new product sequences (similar product set) with higher similarity to the multi-modal representation of the product to be scored, and obtain the corresponding similarity values. That is, after obtaining the top k similar product sets, the key growth performances of the previously new products in the first 10 days and 15 days after entering the field can be counted. For example, exposure, transactions, etc.

[0203] Further, the softmax function is used to convert the above-obtained k similarity values into a probability distribution, ensuring that the values in the output result are between 0 and 1 and their sum is 1 to obtain the weight weight. Finally, the obtained weight is used to perform a weighted sum of the statistical features (such as exposure, transactions, etc.) of the new product set to obtain the new product similarity sequence feature used to characterize the growth of new products. That is, the similarity is normalized to obtain weight, i.e., weight = softmax(sim_score). Further, the obtained weight weight is used to perform a weighted sum of the initial guess of the new product growth performance data of the new product sequence to obtain the new product similarity sequence feature.

[0204] Figure 8 is a schematic diagram for predicting the future performance and maximum potential of new products according to an embodiment of the present application, as Figure 8 shown, multi-modal similarity retrieval is performed on the field new products from the set of similar new products to obtain Product A, Product B, Product C, up to Product F. The growth performance of new products is shown in the figure. pv_10d can be used to represent the growth performance of a new product entering the field for 10 days, and pv_15d can be used to represent the growth performance of a new product entering the field for 15 days. For Product A, pv_10d_A can be used to represent the growth performance of the new product entering the field for 10 days, pv_15d_A can be used to represent the growth performance of the new product entering the field for 15 days, and pv_reward_10d_A can be used to represent the revenue or value of the new product entering the field for 10 days. The same applies to Product B, Product C, and Product F, which will not be specifically described here.

[0205] This embodiment describes the growth rate and leverage ratio change of new products by introducing a new product similarity sequence. After obtaining the growth performance of similar products, the maximum potential of new products can be predicted based on max pooling. pv_1d_max can be used to represent the maximum potential of a new product entering the field for 1 day, pv_3d_max can be used to represent the maximum potential of a new product entering the field for 3 days, and pv_reward_10d_max can be used to represent the revenue or value of a new product entering the field for 10 days. In addition, after obtaining the similarity between the new product and the similar product, the future performance of the new product can be predicted based on average pooling. pv_1d_avg can be used to represent the performance of a new product after entering the field for 1 day, pv_3d_avg can be used to represent the performance of a new product after entering the field for 3 days, and pv_reward_10d_avg can be used to represent the revenue or value of a new product after entering the field for 10 days.

[0206] After obtaining the new product similarity sequence features, the output new product similarity sequence features representing the growth of the new product can be concatenated with other product features (such as static statistical attributes, generalization features, statistical features, multi-modal features, and the similarity sequence of the first guessed old product), and then input into the new product potential estimation model to predict the growth ability or potential of the new product using the new product potential estimation model.

[0207] Figure 9 is a schematic diagram of invoking the new product potential estimation model according to an embodiment of the present application. As Figure 9 shown, after invoking the similar product retrieval module to obtain the new product similarity sequence, the global statistical features and static attribute features are directly input into the Embedding Layer, and the multi-modal features, the first guessed similarity sequence, and the new product similarity sequence are used as the Q, K, and V of the target attention respectively. Among them, Q can be used to represent the query, K can be used to represent the key, and V can be used to represent the value. That is, the product to be scored is used as the query, the retrieved similar product sequence is used as the key, and after calculating the weights using the target attention method, the similar product sequence is merged based on the weights, and the merged result is input into the Embedding Layer together with other features.

[0208] In this embodiment, since the model needs to perform multi-task processing, for example, predicting the success rate of a product and the traffic that a product can obtain. The importance of these tasks for the underlying features varies, and their respective processing logics are also different. Using the Multi-gate Mixture-of-Experts (abbreviated as MMOE) structure can enable the product selection model to activate different expert networks when processing different tasks, improving the multi-task processing ability. Among them, the MMOE structure involves efficiency track statistical features, experts, gates, outputs, and combinations of one or more experts and their gating mechanisms (Towers).

[0209] As Figure 9As shown, Expert 1 can be used to represent an expert. Each expert is an independent sub-model, which is used to process specific types of inputs or solve specific tasks. In MMOE, experts can focus on learning local features of the input data or solving specific sub-problems, which helps to improve the overall performance and efficiency of the model. Gate A, Gate B, and Gate C can be used to represent three gating mechanisms respectively. The gating mechanisms dynamically calculate the weights of each expert based on the input features. Tower A, Tower B, and Tower C can be used to represent the combinations of the corresponding experts and their gating mechanisms respectively. Output A, Output B, and Output C can be used to represent the outputs corresponding to the three gating mechanisms respectively, that is, the output is the final result generated after the model processes the input data.

[0210] This embodiment does not rely on commodity behavior data, and similar sequences of new products can also be retrieved through the multi-modal features of the products themselves. The retrieval candidate set is the new products hatched in this scenario in the past, which is relatively consistent with the distribution of new products. The introduced product-side features are the key node data after the product enters this scenario, which can more accurately describe the growth process of the new product after entering this scenario.

[0211] This embodiment uses a service inference model, which can generate high-quality multi-modal representations of products. By finding the new products hatched in this scenario in the past through multi-modal vectors, and weighting and fusing the features of the new products at the key time nodes after entering this scenario according to the similarity. On the one hand, the function of retrieving similar sequences can be realized without relying on commodity behavior data. On the other hand, the indexed candidate products are the new products in the past with the same distribution as the products to be evaluated. The used product-side features are the features retained in the growth process of the new products, which are more instructive, thus greatly improving the model's ability to predict the growth of new products.

[0212] In the embodiment of the present application, based on the multi-modal information of the target object, candidate objects consistent with the data distribution state of the target object are determined, and then through the feature data of the candidate objects, the feature data for characterizing the performance result of the target object is predicted, avoiding the situation that due to the lack of behavioral data of the target object, the feature data cannot be effectively determined based on the behavioral data of the target object, achieving the purpose of accurately describing the performance result of the target object, and further realizing the technical effect of improving the data generation efficiency of the object, and solving the technical problem of low data generation efficiency of the object.

[0213] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data for analysis, stored data, displayed data, etc.) involved in this application are all information and data that have been authorized by the user or fully authorized by all parties. And the collection, use, and processing of relevant data need to comply with the relevant laws, regulations, and standards of relevant countries and regions, and corresponding operation entrances are provided for users to choose to authorize or refuse.

[0214] It should be noted that for the foregoing method embodiments, for the sake of simple description, they are all expressed as a series of action combinations. However, those skilled in the art should know that this application is not limited by the described action sequence, because according to this application, certain steps can be performed in other sequences or simultaneously. Secondly, those skilled in the art should also know that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily essential to this application.

[0215] Through the description of the above embodiments, those skilled in the art can clearly understand that the method according to the above embodiments can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on such an understanding, the technical solution of this application, in essence, or the part that contributes to the prior art can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk), and includes several instructions for causing a terminal device (which can be a mobile phone, computer, server, or network device, etc.) to execute the methods described in various embodiments of this application.

[0216] According to an embodiment of this application, there is also provided an object feature data generation device for implementing the Figure 2 object feature data generation method for the object features shown above.

[0217] Figure 10 is a schematic diagram of an object feature data generation device according to an embodiment of this application, as Figure 10 shown. The object feature data generation device 1000 may include: a first acquisition unit 1002, a determination unit 1004, a second acquisition unit 1006, and a prediction unit 1008.

[0218] The first acquisition unit 1002 is used to acquire multi-modal information of a target object.

[0219] The determination unit 1004 is used to determine at least one candidate object based on the multi-modal information of the target object, where the similarity between the data distribution state of the candidate object and the data distribution state of the target object exceeds a first threshold.

[0220] A second acquisition unit 1006, configured to acquire feature data of a candidate object, where the feature data of the candidate object is used to represent the performance result of the candidate object on the object recommendation platform.

[0221] A prediction unit 1008, configured to predict feature data of a target object based on the feature data of the candidate object, where the feature data of the target object is used to represent the performance result of the target object on the object recommendation platform.

[0222] It should be noted here that the above first acquisition unit 1002, determination unit 1004, second acquisition unit 1006, and prediction unit 1008 correspond to steps S202 to S208 in the above embodiment. The instances and application scenarios implemented by the four modules and the corresponding steps are the same, but are not limited to the content disclosed in the above embodiment. It should be noted that the above modules or units may be hardware components or software components stored in a memory and processed by one or more processors, and the above modules may also be part of a device and may run in the server 10 provided in the above embodiment.

[0223] In this object feature data generation device, the multimodal information of the target object is acquired by the first acquisition unit 1002. Based on the multimodal information of the target object, the determination unit 1004 determines at least one candidate object, where the similarity between the data distribution state of the candidate object and the data distribution state of the target object exceeds a first threshold. The second acquisition unit 1006 acquires the feature data of the candidate object, where the feature data of the candidate object is used to represent the performance result of the candidate object on the object recommendation platform. The prediction unit 1008 predicts the feature data of the target object based on the feature data of the candidate object, where the feature data of the target object is used to represent the performance result of the target object on the object recommendation platform. That is to say, this object feature data generation device determines a candidate object whose data distribution state is consistent with that of the target object based on the multimodal information of the target object, and then predicts the feature data used to characterize the performance result of the target object through the feature data of the candidate object, avoiding the situation that due to the lack of behavioral data of the target object, the feature data cannot be effectively determined based on the behavioral data of the target object, achieving the purpose of accurately describing the performance result of the target object, and further realizing the technical effect of improving the data generation efficiency of the object, and solving the technical problem of low data generation efficiency of the object.

[0224] According to an embodiment of the present application, there is also provided a Figure 3 feature data generation device for an object for implementing the above

[0225] Figure 11 is a schematic diagram of another object feature data generation device according to an embodiment of the present application, as Figure 11As shown, the feature data generation device 1100 of the object may include: a third acquisition unit 1102, a second determination unit 1104, a fourth acquisition unit 1106, and a second prediction unit 1108.

[0226] The third acquisition unit 1102 is configured to acquire multimodal information of a target transaction object.

[0227] The second determination unit 1104 is configured to determine at least one candidate transaction object based on the multimodal information of the target transaction object, where the similarity between the data distribution state of the candidate transaction object and the data distribution state of the target transaction object exceeds a first threshold.

[0228] The fourth acquisition unit 1106 is configured to acquire feature data of the candidate transaction object, where the feature data of the candidate transaction object is used to represent the performance result of the candidate transaction object on the e-commerce platform.

[0229] The second prediction unit 1108 is configured to predict the feature data of the target transaction object based on the feature data of the candidate transaction object, where the feature data of the target transaction object is used to represent the performance result of the target transaction object on the e-commerce platform.

[0230] It should be noted here that the above-mentioned third acquisition unit 1102, second determination unit 1104, fourth acquisition unit 1106, and second prediction unit 1108 correspond to steps S302 to S308 in the above embodiment. The instances and application scenarios implemented by the four modules and the corresponding steps are the same, but are not limited to the content disclosed in the above embodiment. It should be noted that the above-mentioned module or unit may be a hardware component or a software component stored in a memory and processed by one or more processors. The above-mentioned module may also be part of a device and may run in the server 10 provided in the above embodiment.

[0231] In the feature data generation device of the object, the multimodal information of the target transaction object is obtained by the third acquisition unit 1102. Based on the multimodal information of the target transaction object, at least one candidate transaction object is determined by the second determination unit 1104, where the similarity between the data distribution state of the candidate transaction object and the data distribution state of the target transaction object exceeds the first threshold. The feature data of the candidate transaction object is obtained by the fourth acquisition unit 1106, where the feature data of the candidate transaction object is used to represent the performance result of the candidate transaction object on the e-commerce platform. The feature data of the target transaction object is predicted based on the feature data of the candidate transaction object by the second prediction unit 1108, where the feature data of the target transaction object is used to represent the performance result of the target transaction object on the e-commerce platform. That is to say, the feature data generation device of the object determines a candidate transaction object whose data distribution state is consistent with that of the target transaction object based on the multimodal information of the target transaction object, and then predicts the feature data used to characterize the performance result of the target transaction object through the feature data of the candidate transaction object, avoiding the situation that due to the lack of behavior data of the target transaction object, the feature data cannot be effectively determined based on the behavior data of the target transaction object, achieving the purpose of accurately describing the performance result of the target transaction object, and further realizing the technical effect of improving the data generation efficiency of the object, and solving the technical problem of low data generation efficiency of the object.

[0232] According to an embodiment of the present application, there is also provided an object feature data generation device for implementing the above Figure 4 object feature data generation method shown.

[0233] Figure 12 is a schematic diagram of another object feature data generation device according to an embodiment of the present application, as Figure 12 shown, the object feature data generation device 1200 may include: a first display unit 1202 and a second display unit 1204.

[0234] The first display unit 1202 is configured to display the multimodal information of the target object on the operation interface in response to an input operation on the operation interface.

[0235] The second display unit 1204 is configured to display the feature data of the target object on the operation interface in response to a feature prediction operation on the operation interface, where the feature data of the target object is predicted based on the feature data of at least one candidate object and is used to represent the performance result of the target object on the object recommendation platform, the feature data of the candidate object is used to represent the performance result of the candidate object on the object recommendation platform, the candidate object is determined based on the multimodal information of the target object, and the similarity between the data distribution state of the candidate object and the data distribution state of the target object exceeds the first threshold.

[0236] It should be noted here that the above first display unit 1202 and second display unit 1204 correspond to steps S402 to S404 in the above embodiment. The instances and application scenarios implemented by the two modules and the corresponding steps are the same, but are not limited to the content disclosed in the above embodiment. It should be noted that the above module or unit can be a hardware component or software component stored in the memory and processed by one or more processors. The above module can also be a part of the device and can run in the server 10 provided in the above embodiment.

[0237] In the feature data generation device of the object, the first display unit 1202 responds to an input operation on the operation interface and displays the multimodal information of the target object on the operation interface. The second display unit 1204 responds to a feature prediction operation on the operation interface and displays the feature data of the target object on the operation interface, where the feature data of the target object is predicted based on the feature data of at least one candidate object and is used to represent the performance result of the target object on the object recommendation platform. The feature data of the candidate object is used to represent the performance result of the candidate object on the object recommendation platform. The candidate object is determined based on the multimodal information of the target object, and the similarity between the data distribution state of the candidate object and the data distribution state of the target object exceeds a first threshold. That is to say, the feature data generation device of the object responds to a feature prediction operation on the operation interface and displays the feature data of the target object. The feature data of the target object is predicted based on the feature data of at least one candidate object, and the candidate object is determined based on the multimodal information of the target object and is consistent with the data distribution state of the target object, avoiding the situation that due to the lack of behavior data of the target object, the feature data cannot be effectively determined based on the behavior data of the target object, achieving the purpose of accurately describing the performance result of the target object, and further realizing the technical effect of improving the data generation efficiency of the object and solving the technical problem of low data generation efficiency of the object.

[0238] According to an embodiment of the present application, there is also provided an object feature data generation device for implementing the Figure 5 object feature data generation method shown above.

[0239] Figure 13 is a schematic diagram of another object feature data generation device according to an embodiment of the present application, as Figure 13 shown. The object feature data generation device 1300 may include: a fifth acquisition unit 1302, a third determination unit 1304, a sixth acquisition unit 1306, a third prediction unit 1308, and an output unit 1310.

[0240] The fifth acquisition unit 1302 is configured to acquire multimodal information of a target object by invoking a first interface, where the first interface includes a first parameter, and the parameter value of the first parameter includes the multimodal information of the target object.

[0241] The third determination unit 1304 is configured to determine at least one candidate object based on the multimodal information of the target object, where the similarity between the data distribution state of the candidate object and the data distribution state of the target object exceeds a first threshold.

[0242] The sixth acquisition unit 1306 is configured to acquire feature data of the candidate object, where the feature data of the candidate object is used to represent the performance result of the candidate object on the object recommendation platform.

[0243] The third prediction unit 1308 is configured to predict the feature data of the target object based on the feature data of the candidate object, where the feature data of the target object is used to represent the performance result of the target object on the object recommendation platform.

[0244] The output unit 1310 is configured to output the feature data of the target object by invoking a second interface, where the second interface includes a second parameter, and the parameter value of the second parameter includes the feature data of the target object.

[0245] It should be noted here that the above-mentioned fifth acquisition unit 1302, third determination unit 1304, sixth acquisition unit 1306, third prediction unit 1308, and output unit 1310 correspond to steps S502 to S510 in the above embodiment. The instances and application scenarios implemented by the five modules and the corresponding steps are the same, but are not limited to the content disclosed in the above embodiment. It should be noted that the above modules or units can be hardware components or software components stored in a memory and processed by one or more processors. The above modules can also be part of a device and can run in the server 10 provided in the above embodiment.

[0246] In the feature data generation device of the object, the fifth acquisition unit 1302 acquires the multimodal information of the target object by invoking the first interface. The first interface includes a first parameter, and the parameter value of the first parameter includes the multimodal information of the target object. The third determination unit 1304 determines at least one candidate object based on the multimodal information of the target object, where the similarity between the data distribution state of the candidate object and the data distribution state of the target object exceeds a first threshold. The sixth acquisition unit 1306 acquires the feature data of the candidate object, where the feature data of the candidate object is used to represent the performance result of the candidate object on the object recommendation platform. The third prediction unit 1308 predicts the feature data of the target object based on the feature data of the candidate object, where the feature data of the target object is used to represent the performance result of the target object on the object recommendation platform. The output unit 1310 outputs the feature data of the target object by invoking the second interface. The second interface includes a second parameter, and the parameter value of the second parameter includes the feature data of the target object. That is to say, the feature data generation device of the object determines a candidate object whose data distribution state is consistent with that of the target object based on the multimodal information of the target object, and then predicts the feature data used to characterize the performance result of the target object through the feature data of the candidate object, avoiding the situation that due to the lack of behavioral data of the target object, the feature data cannot be effectively determined based on the behavioral data of the target object, achieving the purpose of accurately describing the performance result of the target object, and further realizing the technical effect of improving the data generation efficiency of the object and solving the technical problem of low data generation efficiency of the object.

[0247] It should be noted that the preferred implementation schemes involved in the above embodiments of this application are the same as the schemes, application scenarios, and implementation processes provided in the above embodiments, but are not limited to the schemes provided in the above embodiments.

[0248] An embodiment of this application can provide a computing device. Figure 14 It is a structural block diagram of a computing device according to an embodiment of this application, as Figure 14 shown. The computing device 1400 may include: one or more (only one is shown in the figure) processors 1402, a memory 1404, a storage controller, and a peripheral interface.

[0249] The above computing device can be understood as an integrated intelligent terminal, including but not limited to servers, desktop computers, personal computers (abbreviated as PC machines), model all-in-ones, etc. And, the model described in the above embodiments of this application may be pre-installed in the computing device.

[0250] Specifically, the computing device can pre - set various types of models, including but not limited to models in the fields of natural language processing, visual processing, speech processing, code processing, multi - modal task processing, etc., so as to provide diverse model selections. In different product forms, the computing device can support one or more model usage methods, including but not limited to model training, model invocation, model fine - tuning, model deployment, model inference and application, etc. In some product forms, the computing device also supports model management, including but not limited to multi - type model management (supporting the management of various types of models such as discriminative and generative models), model version control (supporting the control of different model versions), model evaluation (evaluating the performance and effect of the model based on model evaluation tools), etc. In other product forms, the computing device can also create applications based on the model, provide API invocation capabilities, and can call the model into the created application through the API interface, while providing application management tools to realize the control of the application.

[0251] Furthermore, the computing device can also include data management (supporting the creation and management of model tuning data sets), a training center (providing rich training resources to help users learn and master AI technologies), and basic control capabilities (providing enterprise - level basic control capabilities to ensure the security and efficient operation of the system). Through the above functions, a comprehensive and integrated AI development, training, deployment, and application device is provided.

[0252] Among them, the memory can be used to store software programs and modules, such as the program instructions / modules corresponding to the methods and devices in the embodiments of the present application. The processor executes various functional applications and data processing by running the software programs and modules stored in the memory, that is, implements the methods in the above - mentioned embodiments. The memory can include high - speed random access memory, and can also include non - volatile memory, such as one or more magnetic storage devices, flash memory, or other non - volatile solid - state memories. In some instances, the memory can further include memories remotely set relative to the processor, and these remote memories can be connected to the terminal through a network. Examples of the above - mentioned network include but are not limited to the Internet, enterprise intranets, local area networks, mobile communication networks, and their combinations.

[0253] The processor can call the executable program stored in the memory through the transmission device to execute the method described in any one of the above - mentioned embodiments.

[0254] The embodiments of the present application can provide an electronic device. Figure 15 is a structural block diagram of an electronic device according to an embodiment of the present application. As Figure 15 shown, the electronic device can include: an input / output device 1502; a memory 1504, and a processor 1506, where the processor 1506 is connected to the input / output device 1502 and the memory 1504 through a bus 1508.

[0255] Among them, the memory can be used to store software programs and modules, such as the program instructions / modules corresponding to the methods and devices in the embodiments of the present application. The processor executes various functional applications and data processing by running the software programs and modules stored in the memory, that is, implements the methods in the above embodiments. The memory may include a high-speed random access memory, and may also include a non-volatile memory, such as one or more magnetic storage devices, flash memories, or other non-volatile solid-state memories. In some instances, the memory may further include a memory remotely disposed relative to the processor, and these remote memories may be connected to the terminal through a network. Examples of the above network include but are not limited to the Internet, enterprise intranet, local area network, mobile communication network, and combinations thereof.

[0256] The processor can call the executable program stored in the memory through the transmission device to execute the method described in any one of the above embodiments.

[0257] Those of ordinary skill in the art can understand that the structure shown in the figure is only schematic, and the computing device can also be a smart phone (such as an Android phone, an iOS phone, etc.), a tablet computer, a palm computer, and a mobile Internet device (abbreviated as MID), a PAD and other terminal devices. The figure does not limit the structure of the above computing device. For example, the computing device 1500 may further include more or fewer components (such as a network interface, a display device, etc.) than those shown in the figure, or have a different configuration from that shown in the figure.

[0258] Those of ordinary skill in the art can understand that all or part of the steps in the various methods of the above embodiments can be completed by instructing the relevant hardware of the terminal device through a program, and the program can be stored in a computer-readable storage medium. The storage medium may include: a flash drive, a read-only memory (abbreviated as ROM), a random access memory (abbreviated as RAM), a magnetic disk, or an optical disc, etc.

[0259] The embodiments of the present application further provide a computer-readable storage medium. Optionally, in this embodiment, the above computer-readable storage medium can be used to save the program code executed by the method provided in the above embodiments.

[0260] Optionally, in this embodiment, the above storage medium may be located in the computing device.

[0261] Optionally, in this embodiment, the computer-readable storage medium is set to store an executable program, and when the executable program runs, it controls the device where the computer-readable storage medium is located to execute the method described in any one of the above embodiments.

[0262] An embodiment of the present application also provides a computer program product. Optionally, in this embodiment, the above computer program product may include a computer program, and when the computer program is executed by a processor, the method provided in the above embodiment is implemented.

[0263] An embodiment of the present application also provides a computer program product. Optionally, the above computer program product may include a non-volatile computer-readable storage medium, and the non-volatile computer-readable storage medium may be used to store a computer program, and when the computer program is executed by a processor, the method provided in the above embodiment is implemented.

[0264] An embodiment of the present application also provides a computer program. Optionally, in this embodiment, when the computer program is executed by a processor, the method provided in the above embodiment is implemented.

[0265] In the above embodiments of the present application, the descriptions of the respective embodiments have their own emphases. For parts not detailed in a certain embodiment, reference may be made to the relevant descriptions of other embodiments.

[0266] In the several embodiments provided by the present application, it should be understood that the disclosed technical content can be implemented in other ways. Among them, the device embodiments described above are merely illustrative. For example, the division of the units is only a logical function division. In actual implementation, there may be other division methods. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the displayed or discussed couplings or direct couplings or communication connections to each other can be through some interfaces. The indirect couplings or communication connections of the units or modules can be in an electrical or other form.

[0267] The units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or may be distributed to multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0268] In addition, the functional units in the respective embodiments of the present application can be integrated in a processing unit, or each unit can exist physically alone, or two or more units can be integrated in one unit. The above integrated units can be implemented in the form of hardware or in the form of software functional units.

[0269] When the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of this application. The aforementioned storage medium includes: various media such as USB flash drives, read-only memory ROM, random access memory RAM, mobile hard disks, magnetic disks, or optical discs that can store program codes.

[0270] The above are only the preferred embodiments of this application. It should be noted that for those of ordinary skill in the art of this technology, without departing from the principle of this application, several improvements and refinements can still be made, and these improvements and refinements should also be regarded as the protection scope of this application.

Claims

1. A method for generating characteristic data of an object, characterized in that, Including: Obtain the multimodal information of the target object; Based on the multimodal information of the target object, determine at least one candidate object, where the similarity between the data distribution state of the candidate object and the data distribution state of the target object exceeds a first threshold; Obtain the feature data of the candidate object, where the feature data of the candidate object is used to represent the performance result of the candidate object on the object recommendation platform; Based on the feature data of the candidate object, predict the feature data of the target object, where the feature data of the target object is used to represent the performance result of the target object on the object recommendation platform.

2. The method according to claim 1, wherein The target object is an object published on the object recommendation platform during the current period. Based on the multimodal information of the target object, determining at least one candidate object includes: Determine, from the object set, the candidate objects whose similarity between the multimodal information and the multimodal information of the target object exceeds a second threshold, where the object set includes objects published on the object recommendation platform during the historical period corresponding to the current period, and the similarity between the data distribution states of the objects in the object set and the target object on the object recommendation platform respectively exceeds the first threshold.

3. The method according to claim 2, wherein Determine, from the object set, the candidate objects whose similarity between the multimodal information and the multimodal information of the target object exceeds a second threshold, including: Convert the multimodal information of the target object into a multimodal feature vector of the target object; Retrieve, from the object set, the objects whose similarity between the multimodal feature vector and the multimodal feature vector of the target object exceeds the second threshold, where the multimodal feature vectors of the objects in the object set are obtained by converting the multimodal information of the objects in the object set; Determine the retrieved objects as the candidate objects.

4. The method according to claim 3, wherein Converting the multimodal information of the target object into a multimodal feature vector of the target object includes: Input the multimodal information of the target object into a service inference model, and use the service inference model to infer the multimodal information of the target object to obtain the multimodal feature vector of the target object, where the service inference model is trained using multimodal information samples.

5. The method according to claim 3, wherein Retrieving, from the object set, the objects whose similarity between the multimodal feature vector and the multimodal feature vector of the target object exceeds the second threshold includes: Call a vector retrieval engine to retrieve, from the object set, the objects whose similarity between the multimodal feature vector and the multimodal feature vector of the target object exceeds the second threshold, where the quality index of the multimodal feature vector of the candidate object meets the index threshold.

6. The method according to claim 2, wherein Obtaining the feature data of the candidate object includes: In the historical period, determine the target historical period when the candidate object was published on the object recommendation platform; Obtain the feature sequence of the candidate object within the target historical period, where the feature sequence includes the feature data under different data performance dimensions, and the feature sequence is used to represent the performance result of the candidate object on the object recommendation platform within the target historical period.

7. The method according to claim 6, characterized in that, Based on the feature data of the candidate object, predict the feature data of the target object, including: Based on the feature sequence of the candidate object and the similarity corresponding to the candidate object, predict the feature sequence of the target object, where the similarity corresponding to the candidate object is used to represent the similarity exceeding the second threshold between the multi-modal information of the candidate object and the multi-modal information of the target object, and the feature sequence of the target object is used to represent the performance result of the object on the object recommendation platform within the future period corresponding to the current period.

8. The method according to claim 7, characterized in that Based on the feature sequence of the candidate object and the similarity corresponding to the candidate object, predict the feature sequence of the target object, including: Normalize the similarities corresponding to multiple candidate objects respectively to obtain the weights corresponding to multiple candidate objects, where the weights are used to represent the importance degree of the feature sequence of the corresponding candidate object to the feature sequence of the target object; Use the weights corresponding to multiple candidate objects to merge the feature sequences of multiple candidate objects to obtain the feature sequence of the target object.

9. The method according to claim 8, wherein Use the weights corresponding to multiple candidate objects to merge the feature sequences of multiple candidate objects to obtain the feature sequence of the target object, including: Use the weights corresponding to multiple candidate objects to perform weighted summation on the feature sequences of multiple candidate objects to obtain a summation result; Concatenate the summation result with the feature sequences of at least one auxiliary object to obtain the feature sequence of the target object, where the auxiliary object is an object other than the object set published on the object recommendation platform within the historical period.

10. The method according to claim 9, wherein Concatenate the summation result with the feature sequences of at least one auxiliary object to obtain the feature sequence of the target object, including: Input the summation result and the feature sequences of the auxiliary objects into a feature prediction model, and use the feature prediction model to concatenate the summation result and the feature sequences of the auxiliary objects to obtain the feature sequence of the target object, where the feature prediction model is trained using the feature sequence samples of object samples.

11. A method for generating characteristic data of an object, characterized in that, Include: Obtain the multi-modal information of the target trading object; Based on the multi-modal information of the target trading object, determine at least one candidate trading object, where the similarity between the data distribution state of the candidate trading object and the data distribution state of the target trading object exceeds the first threshold; Obtain the feature data of the candidate trading object, where the feature data of the candidate trading object is used to represent the performance result of the candidate trading object on the e-commerce platform. Based on the characteristic data of the candidate trading object, the characteristic data of the target trading object is predicted, wherein the characteristic data of the target trading object is used to represent the performance result of the target trading object on the e-commerce platform.

12. The method according to claim 11, wherein The method further includes: Based on the characteristic data of the target trading object, the target trading object is displayed in the display area on the e-commerce platform, wherein the display priority of the display area is higher than that of the areas other than the display area on the e-commerce platform.

13. The method according to claim 12, wherein Based on the characteristic data of the target trading object, displaying the target trading object in the display area on the e-commerce platform includes: Determining the value attribute corresponding to the characteristic data of the target trading object; In response to the value attribute satisfying the value attribute threshold, the target trading object is displayed in the display area on the e-commerce platform.

14. A method for generating characteristic data of an object, characterized in that, It includes: In response to an input operation on the operation interface, multimodal information of the target object is displayed on the operation interface; In response to a characteristic prediction operation on the operation interface, the characteristic data of the target object is displayed on the operation interface, wherein the characteristic data of the target object is predicted based on the characteristic data of at least one candidate object and is used to represent the performance result of the target object on the object recommendation platform, the characteristic data of the candidate object is used to represent the performance result of the candidate object on the object recommendation platform, the candidate object is determined based on the multimodal information of the target object, and the similarity between the data distribution state of the candidate object and the data distribution state of the target object exceeds a first threshold.

15. A method for generating characteristic data of an object, characterized in that, It includes: By calling a first interface, multimodal information of the target object is obtained, wherein the first interface includes a first parameter, and the parameter value of the first parameter includes the multimodal information of the target object; Based on the multimodal information of the target object, at least one candidate object is determined, wherein the similarity between the data distribution state of the candidate object and the data distribution state of the target object exceeds a first threshold; Obtaining the characteristic data of the candidate object, wherein the characteristic data of the candidate object is used to represent the performance result of the candidate object on the object recommendation platform; Based on the characteristic data of the candidate object, the characteristic data of the target object is predicted, wherein the characteristic data of the target object is used to represent the performance result of the target object on the object recommendation platform; The characteristic data of the target object is output by calling a second interface, wherein the second interface includes a second parameter, and the parameter value of the second parameter includes the characteristic data of the target object.

16. A characteristic data generation system for an object, characterized in that It includes: A client for transmitting multimodal information of the target object; A server, configured to determine at least one candidate object based on multimodal information of the target object, wherein a similarity between a data distribution state of the candidate object and a data distribution state of the target object exceeds a first threshold; obtain feature data of the candidate object, wherein the feature data of the candidate object is used to represent a performance result of the candidate object on an object recommendation platform; predict the feature data of the target object based on the feature data of the candidate object, wherein the feature data of the target object is used to represent a performance result of the target object on the object recommendation platform; and return the feature data of the target object to the client.

17. A computing device, characterized in that, Comprising: A memory storing an executable program; A processor configured to run the program, wherein when the program runs, it executes the method according to any one of claims 1 to 15.

18. An electronic device, characterized in that, Comprising: A memory storing an executable program; A processor connected to the memory through a bus and configured to run the program, wherein when the program runs, it executes the method according to any one of claims 1 to 15.

19. A computer-readable storage medium, characterized in that, The computer-readable storage medium comprises a stored executable program, wherein when the executable program runs, it controls the device where the computer-readable storage medium is located to execute the method according to any one of claims 1 to 15.

20. A computer program product, characterized in that, Comprising a computer program, which when executed by a processor, implements the method according to any one of claims 1 to 15.

Images