A method, apparatus, electronic device, and storage medium for determining object features

Abstract

This disclosure relates to a method, apparatus, electronic device, and storage medium for determining object features. The method inputs object description information of a sample recommended object into a large language model. The large language model then performs a transformation analysis of the target interaction behavior based on a first preset thought chain prompt text to obtain a transformation reason. This first preset thought chain prompt text prompts the reasoning process for generating the target interaction behavior for the sample recommended object. The transformation reason includes the reasoning result of at least one dimension of this reasoning process. Based on this transformation reason, the sample recommended object is feature-represented to obtain its first object feature. Finally, based on the first object feature of each sample recommended object in the sample recommended object sequence of the sample interaction object, a second object feature of the sample interaction object is determined. This disclosure improves the accuracy of object representation for specific prediction tasks and reduces processing time, thereby improving the training efficiency and accuracy of recommendation probability prediction models for specific prediction tasks.

Description

Technical Field

[0001] This disclosure relates to the field of computer technology, and in particular to a method, apparatus, electronic device, and storage medium for determining object features. Background Technology

[0002] With the development of artificial intelligence technology, in the field of recommendation, recommendation probability prediction models are usually used to predict the probability of a user making a certain decision (such as clicking, buying, liking, commenting, etc.) regarding the object to be recommended (such as a short video, a product, etc.), and then make recommendations to the user based on the predicted probability.

[0003] When training recommendation probability prediction models for specific prediction tasks (such as clicks, purchases, likes, comments, etc.), the related technologies suffer from low training efficiency and poor prediction accuracy on specific prediction tasks because the feature representation methods of sample users and sample recommendation objects (also known as items) are not only time-consuming but also have poor accuracy. Summary of the Invention

[0004] This disclosure provides a method, apparatus, electronic device, and storage medium for determining object features, to at least address the problems in related technologies where, when training recommendation probability prediction models for specific prediction tasks, the feature representation methods for sample users and sample recommended objects are not only time-consuming but also inaccurate, resulting in low model training efficiency and poor prediction accuracy on prediction tasks. The technical solution of this disclosure is as follows:

[0005] According to a first aspect of the present disclosure, a method for determining object features is provided, comprising:

[0006] For any sample recommendation object in the sample recommendation object set, obtain the object description information of the sample recommendation object;

[0007] The object description information is input into a large language model, and the large language model performs a transformation analysis of the target interaction behavior based on a first preset thought chain prompt text to obtain the transformation reason of the sample recommended object corresponding to the target interaction behavior; the first preset thought chain prompt text is used to prompt the reasoning process for generating the target interaction behavior for the sample recommended object; the transformation reason includes the reasoning result of at least one dimension in the reasoning process, and the reasoning result of each dimension represents the performance of the sample recommended object in generating the target interaction behavior in that dimension;

[0008] Based on the conversion reason of the sample recommendation object corresponding to the target interaction behavior, the sample recommendation object is represented by features to obtain the first object feature of the sample recommendation object;

[0009] Based on the first object features of each sample recommendation object in the sample recommendation object sequence within a historical time period, the second object features of the sample interaction object are determined; wherein, the first object features of the sample recommendation object and the second object features of the sample interaction object are used to train a recommendation probability prediction model, and the recommendation probability prediction model is used to predict the probability of generating the target interaction behavior.

[0010] In some exemplary embodiments, the step of representing the sample recommendation object based on the conversion reason corresponding to the target interaction behavior to obtain the first object feature of the sample recommendation object includes:

[0011] The conversion reason corresponding to the target interaction behavior of the sample recommendation object is input into the large language model. The large language model performs keyword extraction processing on the conversion reason according to the preset keyword extraction prompts to obtain the keyword extraction result corresponding to the sample recommendation object. The preset keyword extraction prompts are used to instruct the keyword extraction task to be performed according to the keyword extraction example. The keyword extraction example represents the keywords in the example text output according to the specified format.

[0012] Based on the keyword extraction results corresponding to the sample recommendation object, the first object feature of the sample recommendation object is generated.

[0013] In some exemplary embodiments, generating the first object feature of the sample recommendation object based on the keyword extraction results corresponding to the sample recommendation object includes:

[0014] The conversion reason corresponding to the target interaction behavior of the sample recommendation object is processed by text encoding to obtain the text encoding result corresponding to the sample recommendation object;

[0015] The keyword extraction results corresponding to the sample recommendation object are used as sparse features, and the text encoding results corresponding to the sample recommendation object are used as dense features. The sparse features and the dense features are concatenated to obtain the first object feature of the sample recommendation object.

[0016] In some exemplary embodiments, after inputting the object description information into a large language model, and using the large language model to perform conversion analysis of target interaction behavior based on a first preset thought chain prompt text to obtain the conversion reason of the sample recommended object, the method further includes:

[0017] Obtain the sequence of sample recommendation objects for the sample interaction object within a historical time period, and determine the positive sample recommendation objects and negative sample recommendation objects in the sample recommendation object sequence; the positive sample recommendation object refers to the sample recommendation object that generated the target interaction behavior, and the negative sample recommendation object refers to the sample recommendation object that did not generate the target interaction behavior;

[0018] Based on the conversion reasons of the positive sample recommendation objects corresponding to the target interaction behavior and the conversion reasons of the negative sample recommendation objects corresponding to the target interaction behavior, the preference conversion reasons of the sample interaction objects when generating the target interaction behavior are determined, and the profile features of the sample interaction objects are generated based on the preference conversion reasons of the sample interaction objects.

[0019] The step of determining the second object feature of the sample interaction object based on the first object feature of each sample recommendation object in the sample recommendation object sequence within a historical time period includes:

[0020] The first object features of each sample recommendation object in the sample recommendation object sequence and the profile features of the sample interaction object are concatenated to obtain the second object features of the sample interaction object.

[0021] In some exemplary embodiments, before performing feature representation on the sample recommendation object based on the conversion reason corresponding to the target interaction behavior to obtain the first object feature of the sample recommendation object, the method further includes:

[0022] Based on the reasons for preference conversion stated by each sample interaction object, determine the set of reasons for preference conversion;

[0023] For each sample recommendation object corresponding to the target interaction behavior, the conversion reason is adjusted based on the set of preference conversion reasons to obtain the adjusted conversion reason for each sample recommendation object; the adjusted conversion reason is a subset of the set of preference conversion reasons.

[0024] The adjusted conversion reason corresponding to each of the sample recommendation objects is used as the conversion reason of the sample recommendation object corresponding to the target interaction behavior. The step of performing feature representation on the sample recommendation object based on the conversion reason of the sample recommendation object corresponding to the target interaction behavior is executed to obtain the first object feature of the sample recommendation object.

[0025] In some exemplary embodiments, determining the preference conversion reason of the sample interaction object when generating the target interaction behavior based on the conversion reason of the positive sample recommendation object corresponding to the target interaction behavior and the conversion reason of the negative sample recommendation object corresponding to the target interaction behavior, and generating the profile features of the sample interaction object based on the preference conversion reason of the sample interaction object, includes:

[0026] Obtain a second preset thought chain prompt text; the second preset thought chain prompt text is used to prompt the analysis of the preferences of the sample interaction object when generating the target interaction behavior in multiple steps;

[0027] The conversion reasons of the positive sample recommendation object corresponding to the target interaction behavior and the conversion reasons of the negative sample recommendation object corresponding to the target interaction behavior are input into the large language model. The large language model performs preference analysis of the sample interaction object based on the second preset thinking chain prompt text to obtain the profile features of the sample interaction object.

[0028] The multiple steps include, in sequence: identifying positive sample recommendation objects that generate the target interaction behavior and negative sample recommendation objects that do not generate the target interaction behavior; comparing the conversion reasons of the positive sample recommendation objects with the conversion reasons of the negative sample recommendation objects; analyzing the preference conversion reasons of the sample interaction objects when generating the target interaction behavior based on the differences; and predicting the type and degree of preference of the recommended objects preferred by the sample interaction objects based on the preference conversion reasons.

[0029] In some exemplary embodiments, the reasoning process for generating the target interactive behavior for the sample recommendation object includes reasoning from multiple preset dimensions associated with the target interactive behavior.

[0030] In some exemplary embodiments, obtaining the object description information of the sample recommendation object includes:

[0031] Obtain information on multiple modalities of the sample recommendation object;

[0032] Based on one or more of the information from the multiple modalities, determine the object description information of the sample recommendation object.

[0033] According to a second aspect of the present disclosure, an object feature determination apparatus is provided, comprising:

[0034] The object description acquisition unit is configured to acquire the object description information of any sample recommendation object in the sample recommendation object set.

[0035] The conversion analysis unit is configured to input the object description information into a large language model, and perform conversion analysis of the target interaction behavior based on a first preset thought chain prompt text through the large language model to obtain the conversion reason for the sample recommended object corresponding to the target interaction behavior; the first preset thought chain prompt text is used to prompt the reasoning process for generating the target interaction behavior for the sample recommended object; the conversion reason includes the reasoning result of at least one dimension in the reasoning process, and the reasoning result of each dimension characterizes the performance of the sample recommended object in generating the target interaction behavior in that dimension;

[0036] The first feature representation unit is configured to perform feature representation on the sample recommendation object based on the conversion reason corresponding to the target interaction behavior of the sample recommendation object, so as to obtain the first object feature of the sample recommendation object;

[0037] The second feature representation unit is configured to perform a process of determining the second object feature of the sample interaction object based on the first object feature of each sample recommendation object in the sample recommendation object sequence within a historical time period; wherein the first object feature of the sample recommendation object and the second object feature of the sample interaction object are used to train a recommendation probability prediction model, and the recommendation probability prediction model is used to predict the probability of generating the target interaction behavior.

[0038] In some exemplary embodiments, the first feature representation unit includes:

[0039] The keyword extraction unit is configured to input the conversion reason corresponding to the target interaction behavior of the sample recommendation object into the large language model, and then perform keyword extraction processing on the conversion reason according to the preset keyword extraction prompts through the large language model to obtain the keyword extraction result corresponding to the sample recommendation object; the preset keyword extraction prompts are used to instruct the keyword extraction task to be performed according to the keyword extraction example, and the keyword extraction example represents the keywords in the example text output according to a specified format.

[0040] The first object feature generation unit is configured to generate the first object feature of the sample recommendation object based on the keyword extraction results corresponding to the sample recommendation object.

[0041] In some exemplary embodiments, the first object feature generation unit includes:

[0042] The text encoding unit is configured to perform text encoding processing on the conversion reason of the sample recommendation object corresponding to the target interaction behavior, and obtain the text encoding result corresponding to the sample recommendation object;

[0043] The feature concatenation unit is configured to perform the following operations: taking the keyword extraction result corresponding to the sample recommendation object as a sparse feature and the text encoding result corresponding to the sample recommendation object as a dense feature, and concatenating the sparse feature and the dense feature to obtain the first object feature of the sample recommendation object.

[0044] In some exemplary embodiments, the apparatus further includes:

[0045] The sequence acquisition unit is configured to acquire a sequence of sample recommendation objects for a sample interaction object within a historical time period, and determine positive and negative sample recommendation objects in the sample recommendation object sequence; the positive sample recommendation object refers to the sample recommendation object that generated the target interaction behavior, and the negative sample recommendation object refers to the sample recommendation object that did not generate the target interaction behavior.

[0046] The profile feature generation unit is configured to perform actions based on the conversion reasons of the positive sample recommendation object corresponding to the target interaction behavior and the conversion reasons of the negative sample recommendation object corresponding to the target interaction behavior, to determine the preference conversion reasons of the sample interaction object when generating the target interaction behavior, and to generate profile features of the sample interaction object based on the preference conversion reasons of the sample interaction object;

[0047] Accordingly, the second feature representation unit is specifically configured to perform concatenation of the first object features of each sample recommendation object in the sample recommendation object sequence and the profile features of the sample interaction object to obtain the second object features of the sample interaction object.

[0048] In some exemplary embodiments, the apparatus further includes:

[0049] The preference reason aggregation unit is configured to execute the preference conversion reasons based on each sample interaction object to determine a set of preference conversion reasons;

[0050] The correction unit is configured to execute a conversion reason corresponding to the target interaction behavior for each sample recommendation object, and adjust the conversion reason based on the set of preference conversion reasons to obtain an adjusted conversion reason for each sample recommendation object; the adjusted conversion reason is a subset of the set of preference conversion reasons.

[0051] The first feature representation unit is specifically configured to perform the step of using the adjusted conversion reason corresponding to each sample recommendation object as the conversion reason of the sample recommendation object corresponding to the target interaction behavior, and performing feature representation of the sample recommendation object based on the conversion reason of the sample recommendation object corresponding to the target interaction behavior to obtain the first object feature of the sample recommendation object.

[0052] In some exemplary embodiments, the profile feature generation unit is specifically configured to execute the acquisition of a second preset thought chain prompt text; the second preset thought chain prompt text is used to prompt the analysis of the preferences of the sample interaction object when generating the target interaction behavior according to multiple steps; the conversion reasons of the positive sample recommendation object and the conversion reasons of the negative sample recommendation object are input into the large language model, and the large language model performs the preference analysis of the sample interaction object based on the second preset thought chain prompt text to obtain the profile features of the sample interaction object;

[0053] The multiple steps include, in sequence: identifying positive sample recommendation objects that generate the target interaction behavior and negative sample recommendation objects that do not generate the target interaction behavior; comparing the conversion reasons of the positive sample recommendation objects with the conversion reasons of the negative sample recommendation objects; analyzing the preference conversion reasons of the sample interaction objects when generating the target interaction behavior based on the differences; and predicting the type and degree of preference of the recommended objects preferred by the sample interaction objects based on the preference conversion reasons.

[0054] In some exemplary embodiments, the reasoning process for generating the target interactive behavior for the sample recommendation object includes reasoning from multiple preset dimensions associated with the target interactive behavior.

[0055] In some exemplary embodiments, the object description acquisition unit includes:

[0056] The multimodal information acquisition unit is configured to acquire information on multiple modalities of the sample recommendation object;

[0057] The object description determination unit is configured to execute one or more pieces of information based on the multiple modalities to determine the object description information of the sample recommended object.

[0058] According to a third aspect of the present disclosure, an electronic device is provided, comprising:

[0059] processor;

[0060] Memory used to store the processor's executable instructions;

[0061] The processor is configured to execute the instructions to implement the object feature determination method of the first aspect described above.

[0062] According to a fourth aspect of the present disclosure, a computer-readable storage medium is provided, which, when the instructions in the computer-readable storage medium are executed by a processor of an electronic device, enables the electronic device to perform the object feature determination method of the first aspect described above.

[0063] According to a fifth aspect of the present disclosure, a computer program product is provided, including a computer program that, when executed by a processor, implements the object feature determination method of the first aspect described above.

[0064] This embodiment of the disclosure inputs the object description information of any sample recommendation object into a large language model. The large language model then performs a transformation analysis of the target interaction behavior based on a first preset thought chain prompt text to obtain the transformation reason for the sample recommendation object corresponding to the target interaction behavior. The first preset thought chain prompt text is used to prompt the reasoning process for generating the target interaction behavior for the sample recommendation object. The transformation reason includes the reasoning result of at least one dimension in the reasoning process. The reasoning result of each dimension characterizes the performance of the sample recommendation object in generating the target interaction behavior in that dimension. Then, based on the transformation reason of the sample recommendation object, a feature representation is performed on the sample recommendation object to obtain the first object feature of the sample recommendation object. Finally, based on each sample in the sample recommendation object sequence within a historical time period... The first object feature of the recommended object is used to determine the second object feature of the sample interaction object. Since the above process is a reasoning process for generating the target interaction behavior, i.e., oriented towards a specific prediction task, it can extract deep-level information with information gain and easy generalization, which is closely related to the final prediction task. It can remove a lot of redundant content, improve the accuracy of object representation for specific prediction tasks, and perform logical reasoning of target interaction behavior based only on the object description information of the sample recommended object. It follows the pre-trained embedding paradigm without training, saving a lot of time complexity and reducing the time consumption of object representation. In this way, it improves the efficiency of training the recommendation probability prediction model for estimating the probability of generating the target interaction behavior using the first object feature and the second object feature, as well as the prediction accuracy of the model.

[0065] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and are not intended to limit this disclosure. Attached Figure Description

[0066] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this disclosure and, together with the description, serve to explain the principles of this disclosure, and are not intended to unduly limit this disclosure.

[0067] Figure 1 This is a schematic diagram illustrating an application environment of an object feature determination method according to an exemplary embodiment;

[0068] Figure 2 This is a flowchart illustrating an object feature determination method according to an exemplary embodiment;

[0069] Figure 3 This is a flowchart illustrating another method for determining object features according to an exemplary embodiment;

[0070] Figure 4 This is a flowchart illustrating another method for determining object features according to an exemplary embodiment;

[0071] Figure 5 This is a structural block diagram illustrating an object feature determination device according to an exemplary embodiment;

[0072] Figure 6 This is a structural block diagram of an electronic device according to an exemplary embodiment. Detailed Implementation

[0073] To enable those skilled in the art to better understand the technical solutions of this disclosure, the technical solutions in the embodiments of this disclosure will be clearly and completely described below with reference to the accompanying drawings.

[0074] It should be noted that the terms "first," "second," etc., used in the specification, claims, and accompanying drawings of this disclosure are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this disclosure described herein can be implemented in orders other than those illustrated or described herein. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this disclosure. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this disclosure as detailed in the appended claims.

[0075] It should also 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 used for display, data used for analysis, etc.) involved in this disclosure are all information and data authorized by the user or fully authorized by all parties.

[0076] Please see Figure 1 The diagram illustrates an application environment according to an exemplary embodiment. The application environment may include a terminal 110 and a server 120, which can be connected via a wired network or a wireless network.

[0077] Terminal 110 can be a smartphone, tablet, laptop, desktop computer, etc., but is not limited to these. Terminal 110 may have client software, such as an application (App), installed to provide interactive functionality. This application can be a standalone application or a subroutine within an application. Users of terminal 110 can log in to the application using pre-registered user information, which may include an account and password. For example, the application could be a social networking application, a live streaming application, or a video application.

[0078] Server 120 may be a server providing background services for applications in terminal 110. These background services may include a recommendation service, where the recommended items can be based on items in the actual recommendation system, such as videos, products, images and text, news, and music. Server 120 may store a trained recommendation probability prediction model, which is used to predict the probability of generating a target interactive behavior. This target interactive behavior can characterize the interaction intent of the interactive object. When the interactive object generates a target interactive behavior towards a recommended object, it indicates that the recommended object has achieved a conversion. The target interactive behavior can be any one of the following: clicking, purchasing, downloading, liking, following, or long-running, and the specific behavior can be determined based on actual needs.

[0079] In this embodiment of the disclosure, server 120 may be an independent physical server, a server cluster or distributed system composed of multiple physical servers, an edge computing node, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDN (Content Delivery Network), and big data and artificial intelligence platforms.

[0080] Figure 2 This is a flowchart illustrating an object feature determination method according to an exemplary embodiment, such as... Figure 2 As shown, it includes the following steps:

[0081] In step S201, for any sample recommendation object in the sample recommendation object set, the object description information of the sample recommendation object is obtained.

[0082] The sample recommendation objects in the sample recommendation object set can be used to recommend items to interactive objects for exposure and display on the interactive object's terminal. For example, sample recommendation objects can be internet products such as videos, goods, images and text, news, and music. Interactive objects are those that perform interactive behaviors, which can include browsing, clicking, playing, liking, commenting, and purchasing.

[0083] The object description information of the sample recommendation object is comprehensively described in text form. For example, in order to improve the accuracy of object feature determination, the object description information of the sample recommendation object should describe the sample recommendation object as comprehensively as possible. Therefore, the above step S201 may include: obtaining information on multiple modalities of the sample recommendation object; and determining the object description information of the sample recommendation object based on one or more of the information on the multiple modalities.

[0084] Specifically, multiple modalities can include text, images, and speech. Among them, the information in the text modality can be the text content of the sample recommendation object, such as the title; the information in the image modality can be the image content of the sample recommendation object recognized by image recognition technology (such as OCR, Optical Character Recognition); and the information in the speech modality can be the speech content of the sample recommendation object recognized by automatic speech recognition technology (ASR, Automatic Speech Recognition).

[0085] When determining the object description information of the sample recommendation object based on one or more of the information from these multiple modalities, the sample recommendation object can be described from multiple descriptive dimensions of text. These multiple descriptive dimensions can include modal dimensions and attribute dimensions. For example, modal dimensions can include text, voice, and images, while attribute dimensions can include categories, associated objects, etc. Taking a short video advertisement as an example, these multiple descriptive dimensions can include the video title, the content displayed in the video, the subtitles accompanying the video, the voice accompanying the video, the product recommended in the video, the category to which the product belongs, the price of the product, and the average price of products in the category to which the product belongs.

[0086] In step S203, the object description information of the sample recommended object is input into the large language model. The large language model performs a conversion analysis of the target interaction behavior based on the first preset thought chain prompt text to obtain the conversion reason of the sample recommended object.

[0087] The first preset thought chain prompt text is used to prompt the reasoning process for generating the target interactive behavior for the sample recommended object. The conversion reason includes the reasoning result of at least one dimension in the reasoning process, and the reasoning result of each dimension represents the performance of the sample recommended object in generating the target interactive behavior in that dimension.

[0088] Large language models (LLMs) are based on deep learning and neural network technologies, particularly the Transformer architecture. Their specific functions include natural language understanding, text generation, dialogue systems, language translation, and text summarization. Through training on massive amounts of text data, they can predict contextual relationships and achieve semantic understanding and generation. A chain-of-thought (COT) refers to the logically connected thought processes that form when the model generates its output.

[0089] Specifically, the first preset thought chain prompt text can be input into the large language model to stimulate its creativity and semantic understanding capabilities, guiding it to perform a transformation analysis task of the target interactive behavior. This involves step-by-step reasoning to deduce possible transformation reasons for achieving the target interactive behavior, i.e., the reasons that would lead to the target interactive behavior. These transformation reasons could be, for example, a chain structure like "because a, therefore b; because b, therefore c". Therefore, the transformation reasons of the sample recommended object corresponding to the target interactive behavior can also be called the thought chain of the sample recommended object corresponding to the target interactive behavior. This allows for the extraction of the thought chain of the sample recommended object for specific prediction tasks (i.e., predicting the probability of generating the target interactive behavior). Understandably, the transformation reasons of the sample recommended object corresponding to the target interactive behavior output by the large language model can include content not present in the object description information but inferred by the large language model that is related to generating the target interactive behavior.

[0090] In some examples, the first preset thought chain prompt text can be designed as a closed-loop thought chain pattern. That is, the reasoning process for generating the target interactive behavior for the sample recommendation object includes reasoning from multiple preset dimensions associated with the target interactive behavior. Furthermore, the conversion reason for the sample recommendation object can include the reasoning results of these multiple preset dimensions. For example, these multiple preset dimensions can be general thought chain dimensions for measuring the sample recommendation object in the recommendation task, such as thought chain dimensions that consider effectiveness and superiority—dimensions common to most recommendation tasks.

[0091] Taking e-commerce short video ads as an example, and the target interaction behavior as purchasing, the reasoning based on the aforementioned multiple preset dimensions can be questions with different dimensions such as validity, superiority, applicability, influence, and convenience. In specific implementation, the first preset thought chain prompt text can include instruction text. This instruction text prompts the large language model to perform step-by-step transformation analysis of the target interaction behavior based on the object description information of the input sample recommendation object, following the thought chain approach. For example, the instruction text could be: "Please help me analyze how the following short video e-commerce ad argues to users that they should buy the product promoted by the ad? Please analyze from several aspects: 1. Problem classicity: Is the problem / pain point solved classic and popular? 2. Purchase interest stimulation method: What methods does the ad use to stimulate users' purchase interest? 3. Effectiveness: How does the product effectively solve the user's specific needs? Does it solve problems that traditional methods cannot solve? 4. Price advantage: Is there a price advantage? 5. Convenience: Is the ease of use / convenience good (is there an operation demonstration?)?" (Example) 6. Superiority: Clearly answer whether it involves comparison with competitors? And which competitors are compared? Why is this product superior to competitors in terms of effectiveness and cost? If it doesn't explicitly demonstrate superiority through comparison with competitors, explain how the advertisement implicitly assumes that this product is significantly superior to competitors. 7. Suitability to Consumption Scenarios: Does it introduce specific times, holidays, or consumption / use / enjoyment scenarios that make users feel that using the product is particularly suitable? 8. Scope of Application: Is it applicable to multiple situations and scenarios? 9. Adaptability: Does it have good adaptability to other related products for existing users? 10. Influence: Does it describe user reviews, sales volume, etc.? 11. Brand: What short and memorable words does the advertisement use to make a strong impression on users, such as slogans, repetitive phrases, catchy phrases, and humorous phrases? Let's think step by step, using common sense and knowledge from various disciplines.

[0092] For a specific short video ad, a concrete example of the conversion reason output by the large language model based on the aforementioned first preset thought chain prompt text for purchase conversion analysis could be: "Based on the provided short video e-commerce ad, we can analyze it from the following aspects:

[0093] 1. Problem Classicity: Is the problem / pain point being addressed classic and popular? This advertisement primarily addresses the problems of dandruff control, oil control, and improving the scalp environment. This is a classic and popular problem because dandruff and oily hair are issues faced by many people.

[0094] 2. Methods to Stimulate Purchase Interest: What methods did this advertisement use to stimulate users' purchase interest? The advertisement used XXXX as a spokesperson, whose identity added credibility to the product. Furthermore, the advertisement showcased the product's actual effects, demonstrating how to use the shampoo and its results through a video demonstration.

[0095] 3. Effectiveness: How does this product effectively address the user's specific needs? Does it solve problems that traditional methods cannot? The advertisement emphasizes the product's main ingredient, selenium disulfide, which deeply cleanses hair follicles and improves the scalp environment. This ingredient has a significant effect on improving problems such as dandruff, itchy scalp, and oily scalp. Compared with traditional shampoos, this product is effective and scientifically proven.

[0096] 4. Price Advantage: Does it have a price advantage? The price of this product is Y yuan, which has a significant price advantage compared to the average price.

[0097] 5. Convenience: Is it easy to use / convenient (is there a demonstration)? The advertisement shows a demonstration of how to use the shampoo, making it easier for users to understand how to use the product.

[0098] 6. Superiority: Clearly answer whether a comparison with competitors is involved? And which competitors are compared? Why is this product superior to competitors in terms of effectiveness and cost? The advertisement does not explicitly mention competitors, but by emphasizing the product's effectiveness and price advantage, it implies that the product is superior to competitors.

[0099] 7. Suitability to Consumption Scenarios: Does the advertisement mention specific times, holidays, or consumption / use / enjoyment scenarios that make users feel particularly suitable for using the product? The advertisement does not mention specific times, holidays, or consumption scenarios, but the shampoo and conditioner are suitable for daily use.

[0100] 8. Scope of application: Is it suitable for multiple situations and scenarios? This product is suitable for people with problems such as dandruff, itchy scalp, and oily scalp, and is suitable for multiple situations.

[0101] 9. Compatibility: Does the product have good compatibility with other related products for existing users? The advertisement does not mention the product's compatibility with other related products.

[0102] 10. Impact: Does the advertisement describe user reviews, sales figures, etc.? The advertisement mentions authoritative efficacy testing reports for the product, increasing its credibility. However, it does not mention specific sales figures or users.

[0103] In the above embodiments, since the reasoning process for generating the target interactive behavior for the sample recommendation object includes reasoning from multiple preset dimensions associated with the target interactive behavior, the cost-effectiveness of the demand can be comprehensively addressed, and the information gain of the output conversion reason can be improved.

[0104] In other examples, the first pre-set thought chain prompt text can be designed as an open-ended thought chain model, leveraging the creativity of the large language model to generate open-ended conversion reasons for the sample recommendation object. In specific implementations, this first pre-set thought chain prompt text can include instruction text. This instruction text prompts the large language model to perform step-by-step conversion analysis of the target interaction behavior based on the object description information of the input sample recommendation object, following a thought chain approach. For example, the instruction text could be: "Please help me analyze how the following short video e-commerce advertisement argues with users that they should buy the product promoted in the advertisement? Please explain how each point ultimately leads to a purchase decision. Let's think step by step, using common sense and knowledge from various disciplines."

[0105] For a specific short video ad, a concrete example of the conversion reason output by the large language model based on the aforementioned first preset thought chain prompt text for purchase conversion analysis could be: "This short video e-commerce ad argues to users the reasons for purchasing this shampoo from the following aspects:"

[0106] 1. XXXX endorsement: By recommending XXXX, the credibility of the advertisement is enhanced, making it easier for users to believe in the product's effectiveness.

[0107] 2. Shampoo ingredients: The advertisement mentions "selected golden ratio of selenium disulfide", which lets users know that the product contains effective ingredients, thereby increasing their willingness to buy.

[0108] 3. Deep cleansing of hair follicles: The advertisement emphasizes the function of "deep cleansing of hair follicles" to make users believe that this shampoo can effectively improve the scalp environment.

[0109] 4. Oil control and dandruff removal: The advertisement repeatedly emphasizes the "oil control and dandruff removal" effect, making users believe that this shampoo can solve common problems such as oily scalp, dandruff, and itchy scalp.

[0110] 5. Domestic Brands: By mentioning "domestic brands," advertisements associate products with national pride, making users more willing to support domestic products.

[0111] 6. Demonstration of actual results: The advertisement shows a woman using shampoo in the bathroom, allowing users to visually see the product's effects.

[0112] 7. Price Comparison: The advertisement compares the product price with similar products, showing that the product is reasonably priced and offers good value for money.

[0113] In practical applications, the reasoning process for generating the target interactive behavior based on the sample recommendation object can also include assigning a reasoning role to the large language model. This prompts the large language model to capture the professional knowledge possessed by the reasoning role and to perform reasoning based on this professional knowledge during the reasoning process, thereby improving the professionalism and accuracy of the conversion analysis of the target interactive behavior. For example, a reasoning role can be assigned to the large language model through textual guidance such as "You are a professional short video e-commerce advertising analyst, proficient in sales, photography, short video production skills, marketing psychology, and logical argumentation."

[0114] In step S205, the sample recommendation object is represented by features based on the conversion reason corresponding to the target interaction behavior, and the first object feature of the sample recommendation object is obtained.

[0115] In step S207, the second object feature of the sample interaction object is determined based on the first object feature of each sample recommendation object in the sample recommendation object sequence within the historical time period.

[0116] The first object feature of the sample recommendation object and the second object feature of the sample interaction object are used to train a recommendation probability prediction model, which is used to predict the probability of generating the target interaction behavior.

[0117] The historical time period can be set based on actual needs, such as the most recent week or the most recent month.

[0118] Specifically, each sample recommendation object in the sample recommendation object sequence can be represented by its corresponding first object feature to obtain a first object feature sequence, which is then used as the second object feature of the sample interaction object.

[0119] Because the first preset thought chain prompt text in this embodiment is used to prompt the reasoning process for generating the target interaction behavior for the sample recommendation object, i.e., for a specific prediction task, and the output conversion reason includes the reasoning result of at least one dimension in the reasoning process, with each dimension's reasoning result representing the correlation between the sample recommendation object and the target interaction behavior in that dimension, it is possible to extract deep-level information with information gain and easy generalization, which is closely related to the final prediction task. This removes a large amount of redundant content, improves the accuracy of object representation, and performs logical reasoning for the target interaction behavior based solely on the sample recommendation object, mimicking the pre-trained embedding paradigm without requiring training, saving a significant amount of time complexity and reducing the time consumption of object representation. This, in turn, improves the efficiency and prediction accuracy of the recommendation probability prediction model trained using the first object features and the second object features to predict the probability of generating the target interaction behavior.

[0120] In some exemplary embodiments, to further improve the accuracy of object features and thus enhance the training effect of the probability prediction model, step S205 above, when representing the sample recommendation object based on the conversion reason corresponding to the target interaction behavior to obtain the first object feature of the sample recommendation object, may include:

[0121] The conversion reason corresponding to the target interaction behavior of the sample recommendation object is input into the large language model. The large language model performs keyword extraction processing on the conversion reason according to the preset keyword extraction prompts to obtain the keyword extraction result corresponding to the sample recommendation object. The preset keyword extraction prompts are used to instruct the keyword extraction task to be performed according to the keyword extraction example. The keyword extraction example represents the keywords in the example text output according to the specified format.

[0122] Based on the keyword extraction results corresponding to the sample recommendation object, the first object feature of the sample recommendation object is generated.

[0123] Specifically, based on the summarization function of the large language model and the preset keyword extraction prompts of this embodiment, the large language model can output corresponding keyword extraction results based on the keyword extraction examples in the preset keyword extraction prompts. This can further filter out redundant information and retain key information related to generating the target interactive behavior. As a result, the first object feature of the sample recommendation object generated based on the keyword extraction results and the second object feature based on the first object feature are more matched to the specific prediction task of the recommendation probability prediction model (i.e., predicting the probability of generating the target interactive behavior). This is beneficial to improving the training effect of the recommendation probability prediction model trained based on the first object feature and the second object feature for the specific prediction task.

[0124] In some exemplary embodiments, to further improve the accuracy of object representation for specific prediction tasks and thus enhance the training effect of recommendation probability prediction models for specific prediction tasks, generating the first object feature of the sample recommendation object based on the keyword extraction results corresponding to the sample recommendation object may include:

[0125] The conversion reason corresponding to the target interaction behavior of the sample recommendation object is processed by text encoding to obtain the text encoding result corresponding to the sample recommendation object;

[0126] The keyword extraction results corresponding to the sample recommendation object are used as sparse features, and the text encoding results corresponding to the sample recommendation object are used as dense features. The sparse features and the dense features are concatenated to obtain the first object feature of the sample recommendation object.

[0127] Specifically, a pre-trained language representation model such as the BERT model can be used to perform the above text encoding processing. That is, the conversion reason of the sample recommendation object corresponding to the target interaction behavior is input into the BERT model for text encoding processing to obtain the text encoding result output by the BERT model.

[0128] After obtaining the keyword extraction results and text encoding results of the sample recommendation object, the accuracy of the first object feature can be further improved by concatenating the keyword extraction results as sparse features and the text encoding results as dense features, and using the concatenated result as the first object feature of the sample recommendation object.

[0129] In some exemplary implementations, such as Figure 3 A flowchart illustrating another method for determining object features is provided. In step S203, the object description information is input into a large language model. After the large language model performs a conversion analysis of the target interaction behavior based on the first preset thought chain prompt text to obtain the conversion reason of the sample recommended object, the method may further include:

[0130] In step S301, the sequence of sample recommendation objects for sample interaction objects within a historical time period is obtained, and the positive and negative sample recommendation objects in the sequence of sample recommendation objects are determined.

[0131] Here, the sample interaction object can be any sample interaction object in the set of sample interaction objects. The sample recommendation object sequence within a historical time period refers to the sample recommendation objects recommended to the sample interaction object within the historical time period, such as the sample recommendation objects exposed on the sample interaction object's terminal. The sample interaction object can perform interactive behaviors on the exposed sample recommendation objects, such as browsing, clicking, purchasing, downloading, etc.

[0132] Determining the positive and negative recommended objects in the sequence of recommended objects can be achieved by generating corresponding tag information for each recommended object. This tag information indicates whether the interactive object has generated a target interaction behavior with the corresponding recommended object. For example, the tag information can be "1" or "0", where "1" indicates that the interactive object has generated a target interaction behavior with the recommended object, and "0" indicates that the interactive object has not generated a target interaction behavior with the recommended object. Taking a short video advertisement as an example, if the interactive object purchases a product associated with the short video advertisement based on the exposure of the short video advertisement, the tag information for the short video advertisement is "1"; conversely, if the interactive object does not purchase a product associated with the short video advertisement based on the exposure of the short video advertisement, the tag information for the short video advertisement is "0".

[0133] In this context, a positive sample recommendation object in the sample recommendation object sequence refers to a sample recommendation object that has generated the target interaction behavior, such as a sample recommendation object with a label of "1"; a negative sample recommendation object in the sample recommendation object sequence refers to a sample recommendation object that has not generated the target interaction behavior, such as a sample recommendation object with a label of "0".

[0134] In step S303, based on the conversion reasons of the positive sample recommendation object corresponding to the target interaction behavior and the conversion reasons of the negative sample recommendation object corresponding to the target interaction behavior, the preference conversion reason of the sample interaction object when generating the target interaction behavior is determined, and the profile features of the sample interaction object are generated based on the preference conversion reason of the sample interaction object.

[0135] Specifically, the reason for preference conversion when a sample interaction object generates a target interaction behavior can be a conversion reason belonging to a positive sample recommendation object corresponding to the target interaction behavior, but not a conversion reason belonging to a negative sample recommendation object corresponding to the target interaction behavior. For example, if the sample recommendation object sequence for sample interaction object User1 is {(item1,1), (item2,0), (item3,0)}, then item1 is a positive sample recommendation object, and items2 and ite3 are negative sample recommendation objects. Assuming that the conversion reason for item1 corresponding to the target interaction behavior is (R1,R2), the conversion reason for item2 corresponding to the target interaction behavior is (R2,R3), and the conversion reason for item3 corresponding to the target interaction behavior is (R4,R5), then the preference conversion reason for sample interaction object User1 when generating the target interaction behavior is R1.

[0136] The profile features of the sample interaction object may include the reason for its preference conversion, as well as other information related to the reason for its preference conversion, such as the type of recommended object associated with the reason for its preference conversion.

[0137] Based on this, the aforementioned step S207 can be implemented as follows: Figure 3 In step S305 shown, the first object features of each sample recommendation object in the sample recommendation object sequence and the profile features of the sample interaction object are concatenated to obtain the second object features of the sample interaction object.

[0138] Specifically, each sample recommendation object in the sample recommendation object sequence of the sample interaction object is represented by its corresponding first object feature, and then the portrait features of the sample interaction object are concatenated to obtain the second object feature of the sample interaction object.

[0139] The above implementation determines the reason for the preference conversion of a sample interaction object when generating a target interaction behavior by using the conversion reasons of positive and negative sample recommendation objects in the sample recommendation object sequence based on the sample interaction object. Based on this, a profile feature of the sample interaction object is generated. Then, the second object feature of the sample interaction object is obtained by combining the profile feature and the first object feature of each sample recommendation object in the sample recommendation object sequence. This improves the accuracy of the representation of sample interaction objects for specific prediction tasks, and thus helps to improve the training effect of the recommendation probability prediction model for specific prediction tasks.

[0140] In some exemplary embodiments, in order to further save the time spent training the probability prediction model and improve the model training efficiency, the reasoning ability of the large language model can be used to implement the aforementioned step S303. Specifically, step S303 may include the following when implemented:

[0141] Obtain a second preset thought chain prompt text, which is used to prompt the analysis of the preferences of the sample interaction object when generating the target interaction behavior in multiple steps. These multiple steps include: determining the positive sample recommendation objects that generate the target interaction behavior and the negative sample recommendation objects that do not generate the target interaction behavior; comparing the conversion reasons of the positive sample recommendation objects with the conversion reasons of the negative sample recommendation objects; analyzing the preference conversion reasons of the sample interaction object when generating the target interaction behavior based on the differences; and predicting the type and degree of preference of the recommended object preferred by the sample interaction object based on the preference conversion reasons.

[0142] The conversion reasons corresponding to the target interaction behavior for the positive sample recommendation object and the conversion reasons corresponding to the target interaction behavior for the negative sample recommendation object are input into the large language model. The large language model performs preference analysis on the sample interaction object based on the second preset thought chain prompt text to obtain the profile features of the sample interaction object.

[0143] Specifically, when the large language model performs preference analysis on the sample interaction object based on the above-mentioned second preset thinking chain prompt text, it can reason step by step according to the above multiple steps and output the reasoning results of each step. Thus, the reasoning results of these multiple steps can be used as the profile features of the sample interaction object.

[0144] The above implementation method allows the LLM to comprehensively judge why the conversion reasons of positive sample recommended objects corresponding to target interaction behaviors are more reasonable, and allows the LLM to summarize the thought chain that users are more likely to convert. This not only improves the accuracy of reasoning and makes the profile features of sample interaction objects more accurate, but also is equivalent to training the model based on pre-trained embeddings without training, saving a lot of time complexity and improving the training efficiency and training effect of probability prediction models.

[0145] Taking the example of short video ads as the recommended sample and purchase as the target interaction, the second pre-set thought chain prompt text could be: "Please help me analyze the following short video e-commerce ads. For a specific user A, he purchased the product from ad 1 but not the product from ad 2. Please first clearly answer which ads' products the user purchased and which ads' products he did not. Please compare the differences between ads with and without purchases, and rethink from a higher level, analyzing which argumentation methods and viewpoints the user is inherently more likely to accept in making purchase decisions. Finally, please guess the set of brand archetypes the user prefers and the probability of each archetype (list concisely). Let's think step by step, using common sense and world knowledge."

[0146] In specific implementation, the second preset thought chain prompt text can also include assigning a reasoning role to the large language model, thereby prompting the large language model to capture the professional knowledge possessed by the reasoning role. Based on this professional knowledge, reasoning is performed in preference analysis to improve the professionalism and accuracy of the analysis of the preferences of sample interaction objects when generating target interaction behaviors. For example, the second preset thought chain prompt text can guide the large language model to assign a reasoning role through the following text: "You are a professional short video e-commerce advertising analyst, proficient in sales, photography, short video production skills, marketing psychology, and logical argumentation."

[0147] For example, the second preset thought chain prompt text may also include multiple preset recommended object types, thereby guiding the large language model to predict from these multiple preset recommended object types when predicting the recommended object type of the sample interaction object preference based on the reason for preference conversion. The degree of preference can be represented by probability.

[0148] In some exemplary implementations, to further improve the accuracy of object representation and the training efficiency of the probability prediction model, such as Figure 4 The flowchart of another object feature determination method shown above, before step S205, in which the sample recommendation object is feature-represented based on the conversion reason corresponding to the target interaction behavior to obtain the first object feature of the sample recommendation object, may further include:

[0149] In step S401, a set of preference conversion reasons is determined based on the preference conversion reasons of each sample interaction object.

[0150] Specifically, based on the aforementioned steps S301 to S303, the preference conversion reasons for each sample interaction object in the sample interaction object set can be obtained. Aggregating these preference conversion reasons yields a set of preference conversion reasons. For example, if the preference conversion reason for sample interaction object User1 when generating the target interaction behavior is {R1}, the preference conversion reason for sample interaction object User2 when generating the target interaction behavior is {R2, R3}, and the preference conversion reason for sample interaction object User3 when generating the target interaction behavior is {R2, R4}, then the set of preference conversion reasons is {R1, R2, R3, R4}. Assuming the conversion reason for sample recommendation object item1 is {R1, R3, R5}, then the adjusted conversion reason for sample recommendation object item1 is {R1, R3}. This means removing the conversion reasons for sample recommendation object item1 that do not belong to the set of preference conversion reasons, leaving the adjusted conversion reason for sample recommendation object item1.

[0151] In step S403, for each sample recommendation object corresponding to the conversion reason of the target interaction behavior, the conversion reason is adjusted based on the set of preference conversion reasons to obtain the adjusted conversion reason corresponding to each sample recommendation object.

[0152] In this context, the adjusted conversion reason corresponding to each sample recommendation object is a subset of the set of preference conversion reasons. Therefore, when performing the aforementioned step S205, the adjusted conversion reason corresponding to each sample recommendation object is used as the conversion reason for the sample recommendation object corresponding to the target interaction behavior. The step of performing feature representation on the sample recommendation object based on the conversion reason for the sample recommendation object corresponding to the target interaction behavior is performed to obtain the first object feature of the sample recommendation object.

[0153] The above implementation method adjusts the conversion reasons corresponding to the target interaction behavior of each sample recommendation object by combining the set of preference conversion reasons, thereby improving the fit between the adjusted conversion reasons and the target interaction behavior, improving the accuracy of object representation for specific prediction tasks, and helping to improve the accuracy of recommendation probability prediction models for specific prediction tasks.

[0154] This disclosure, in terms of depth, can extract the reasons for the transformation from specific multimodal information to the displayed information, and further to the underlying reasons for the transformation, making the information increasingly essential, stable, and generalizable step by step. In terms of breadth, this thought chain can uncover various rational, irrational, and subtextual information from the text, taking a significant step forward in breadth. These reasons all encompass the three major systems of the user's mind: the symbolic, the meaning, and the value system. This disclosure is applicable to any "creation from nothing" recommendation system. Furthermore, this disclosure can save significant time complexity in training recommendation probability prediction models for specific prediction tasks.

[0155] Figure 5 This is a structural block diagram illustrating an object feature determination device according to an exemplary embodiment. (Refer to...) Figure 5 The object feature determining device 500 includes:

[0156] The object description acquisition unit 510 is configured to acquire the object description information of any sample recommendation object in the sample recommendation object set.

[0157] The conversion analysis unit 520 is configured to input the object description information into a large language model, and perform conversion analysis of the target interaction behavior based on a first preset thought chain prompt text through the large language model to obtain the conversion reason for the sample recommended object corresponding to the target interaction behavior; the first preset thought chain prompt text is used to prompt the reasoning process for generating the target interaction behavior for the sample recommended object; the conversion reason includes the reasoning result of at least one dimension in the reasoning process, and each reasoning result of the dimension represents the performance of the sample recommended object in generating the target interaction behavior in that dimension;

[0158] The first feature representation unit 530 is configured to perform feature representation on the sample recommendation object based on the conversion reason corresponding to the target interaction behavior of the sample recommendation object, and obtain the first object feature of the sample recommendation object;

[0159] The second feature representation unit 540 is configured to perform a process based on the first object feature of each sample recommendation object in the sample recommendation object sequence within a historical time period to determine the second object feature of the sample interaction object; wherein the first object feature of the sample recommendation object and the second object feature of the sample interaction object are used to train a recommendation probability prediction model, and the recommendation probability prediction model is used to predict the probability of generating the target interaction behavior.

[0160] In some exemplary embodiments, the first feature representation unit 530 includes:

[0161] The keyword extraction unit is configured to input the conversion reason corresponding to the target interaction behavior of the sample recommendation object into the large language model, and then perform keyword extraction processing on the conversion reason according to the preset keyword extraction prompts to obtain the keyword extraction result corresponding to the sample recommendation object; the preset keyword extraction prompts are used to instruct the keyword extraction task to be performed according to the keyword extraction example, and the keyword extraction example represents the keywords in the example text output according to a specified format.

[0162] The first object feature generation unit is configured to generate the first object feature of the sample recommendation object based on the keyword extraction results corresponding to the sample recommendation object.

[0163] In some exemplary embodiments, the first object feature generation unit includes:

[0164] The text encoding unit is configured to perform text encoding processing on the conversion reason corresponding to the target interaction behavior of the sample recommendation object, and obtain the text encoding result corresponding to the sample recommendation object;

[0165] The feature concatenation unit is configured to perform the following operations: using the keyword extraction result corresponding to the sample recommendation object as a sparse feature and the text encoding result corresponding to the sample recommendation object as a dense feature, concatenating the sparse feature and the dense feature to obtain the first object feature of the sample recommendation object.

[0166] In some exemplary embodiments, the device 500 further includes:

[0167] The sequence acquisition unit is configured to acquire a sequence of sample recommendation objects for a sample interaction object within a historical time period, and determine the positive and negative sample recommendation objects in the sequence. The positive sample recommendation object refers to the sample recommendation object that generated the target interaction behavior, and the negative sample recommendation object refers to the sample recommendation object that did not generate the target interaction behavior.

[0168] The profile feature generation unit is configured to perform the following: based on the conversion reasons of the positive sample recommendation object corresponding to the target interaction behavior and the conversion reasons of the negative sample recommendation object corresponding to the target interaction behavior, determine the preference conversion reasons of the sample interaction object when generating the target interaction behavior, and generate profile features of the sample interaction object based on the preference conversion reasons of the sample interaction object;

[0169] Accordingly, the second feature representation unit 540 is specifically configured to perform the concatenation of the first object feature of each sample recommended object in the sample recommended object sequence and the profile feature of the sample interaction object to obtain the second object feature of the sample interaction object.

[0170] In some exemplary embodiments, the device 500 further includes:

[0171] The preference reason aggregation unit is configured to execute the preference conversion reason based on each sample interaction object and determine the set of preference conversion reasons;

[0172] The correction unit is configured to execute a conversion reason corresponding to the target interaction behavior for each sample recommendation object, and adjust the conversion reason based on the set of preference conversion reasons to obtain an adjusted conversion reason for each sample recommendation object; the adjusted conversion reason is a subset of the set of preference conversion reasons.

[0173] The first feature representation unit 530 is specifically configured to perform the following steps: taking the adjusted conversion reason corresponding to each sample recommendation object as the conversion reason for the sample recommendation object corresponding to the target interaction behavior, performing feature representation on the sample recommendation object based on the conversion reason for the sample recommendation object corresponding to the target interaction behavior, and obtaining the first object feature of the sample recommendation object.

[0174] In some exemplary embodiments, the profile feature generation unit is specifically configured to execute the acquisition of a second preset thought chain prompt text; the second preset thought chain prompt text is used to prompt the analysis of the preferences of the sample interaction object when generating the target interaction behavior according to multiple steps; the conversion reasons of the positive sample recommendation object corresponding to the target interaction behavior and the conversion reasons of the negative sample recommendation object corresponding to the target interaction behavior are input into the large language model, and the large language model performs preference analysis of the sample interaction object based on the second preset thought chain prompt text to obtain the profile features of the sample interaction object;

[0175] The multiple steps include, in sequence: identifying positive sample recommendation objects that generate the target interaction behavior and negative sample recommendation objects that do not generate the target interaction behavior; comparing the conversion reasons of the positive sample recommendation objects with the conversion reasons of the negative sample recommendation objects; analyzing the preference conversion reasons of the sample interaction object when generating the target interaction behavior based on the difference; and predicting the type and degree of preference of the sample interaction object based on the preference conversion reasons.

[0176] In some exemplary implementations, the reasoning process for generating the target interactive behavior for the sample recommendation object includes reasoning from multiple preset dimensions associated with the target interactive behavior.

[0177] In some exemplary embodiments, the object description acquisition unit 510 includes:

[0178] The multimodal information acquisition unit is configured to acquire information on multiple modalities of the recommended object in the sample.

[0179] The object description determination unit is configured to execute one or more pieces of information based on the multiple modalities to determine the object description information of the sample recommended object.

[0180] Regarding the apparatus in the above embodiments, the specific manner in which each module performs its operation has been described in detail in the embodiments related to the method, and will not be elaborated upon here.

[0181] In one exemplary embodiment, an electronic device is also provided, including a processor; a memory for storing processor-executable instructions; wherein, when the processor is configured to execute the instructions stored in the memory, it implements the object feature determination method provided in any of the above embodiments.

[0182] The electronic device can be a terminal, a server, or a similar computing device. Taking a server as an example... Figure 6 This is a block diagram illustrating an electronic device performing an object feature determination method according to an exemplary embodiment, such as... Figure 6 As shown, the server 600 can vary significantly due to different configurations or performance. It may include one or more central processing units (CPUs) 610 (CPUs 610 may include, but are not limited to, microprocessors such as MCUs or programmable logic devices such as FPGAs), a memory 630 for storing data, and one or more storage media 620 (e.g., one or more mass storage devices) for storing application programs 623 or data 622. The memory 630 and storage media 620 may be temporary or persistent storage. The program stored in the storage media 620 may include one or more modules, each module may include a series of instruction operations on the server. Furthermore, the CPU 610 may be configured to communicate with the storage media 620 and execute the series of instruction operations stored in the storage media 620 on the server 600. Server 600 may also include one or more power supplies 660, one or more wired or wireless network interfaces 650, one or more input / output interfaces 640, and / or one or more operating systems 621, such as Windows Server™, Mac OS X™, Unix™, Linux™, FreeBSD™, etc.

[0183] The input / output interface 640 can be used to receive or send data via a network. Specific examples of the network described above may include a wireless network provided by the communication provider of server 600. In one example, the input / output interface 640 includes a network interface controller (NIC), which can connect to other network devices via a base station to communicate with the Internet. In another example, the input / output interface 640 may be a radio frequency (RF) module for wireless communication with the Internet.

[0184] Those skilled in the art will understand that Figure 6 The structure shown is for illustrative purposes only and does not limit the structure of the aforementioned electronic device. For example, server 600 may also include... Figure 6 The more or fewer components shown, or having the same Figure 6 The different configurations shown.

[0185] In one exemplary embodiment, a computer-readable storage medium is also provided, which, when the instructions in the computer-readable storage medium are executed by a processor of an electronic device, enables the electronic device to perform the object feature determination method provided in any of the above embodiments.

[0186] In one exemplary embodiment, a computer program product is also provided, including a computer program that, when executed by a processor, implements the object feature determination method provided in any of the above embodiments.

[0187] Other embodiments of this disclosure will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of this disclosure that follow the general principles of this disclosure and include common knowledge or customary techniques in the art not disclosed herein. The specification and examples are to be considered exemplary only, and the true scope and spirit of this disclosure are indicated by the following claims.

[0188] It should be understood that this disclosure is not limited to the precise structures described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of this disclosure is limited only by the appended claims.

Claims

1. A method for determining object features, characterized in that, include: For any sample recommendation object in the sample recommendation object set, obtain the object description information of the sample recommendation object; The object description information is input into a large language model, and the large language model performs a transformation analysis of the target interaction behavior based on the first preset thought chain prompt text to obtain the transformation reason of the sample recommended object corresponding to the target interaction behavior; the first preset thought chain prompt text is used to prompt the reasoning process of generating the target interaction behavior for the sample recommended object; The reason for conversion includes the reasoning result of at least one dimension in the reasoning process, and the reasoning result of each dimension represents the performance of the sample recommendation object in that dimension in generating the target interactive behavior; Based on the conversion reason of the sample recommendation object corresponding to the target interaction behavior, the sample recommendation object is represented by features to obtain the first object feature of the sample recommendation object; Based on the first object features of each sample recommendation object in the sample recommendation object sequence within a historical time period, the second object features of the sample interaction object are determined; wherein, the first object features of the sample recommendation object and the second object features of the sample interaction object are used to train a recommendation probability prediction model, and the recommendation probability prediction model is used to predict the probability of generating the target interaction behavior.

2. The method according to claim 1, characterized in that, The step of representing the sample recommendation object based on the conversion reason corresponding to the target interaction behavior, and obtaining the first object feature of the sample recommendation object, includes: The conversion reason corresponding to the target interaction behavior of the sample recommendation object is input into the large language model. The large language model performs keyword extraction processing on the conversion reason according to the preset keyword extraction prompts to obtain the keyword extraction result corresponding to the sample recommendation object. The preset keyword extraction prompts are used to instruct the keyword extraction task to be performed according to the keyword extraction example. The keyword extraction example represents the keywords in the example text output according to the specified format. Based on the keyword extraction results corresponding to the sample recommendation object, the first object feature of the sample recommendation object is generated.

3. The method according to claim 2, characterized in that, The step of generating the first object feature of the sample recommendation object based on the keyword extraction results corresponding to the sample recommendation object includes: The conversion reason corresponding to the target interaction behavior of the sample recommendation object is processed by text encoding to obtain the text encoding result corresponding to the sample recommendation object; The keyword extraction results corresponding to the sample recommendation object are used as sparse features, and the text encoding results corresponding to the sample recommendation object are used as dense features. The sparse features and the dense features are concatenated to obtain the first object feature of the sample recommendation object.

4. The method according to claim 1, characterized in that, After inputting the object description information into a large language model, and using the large language model to perform conversion analysis of target interaction behavior based on the first preset thought chain prompt text to obtain the conversion reason of the sample recommended object, the method further includes: Obtain the sequence of sample recommendation objects for the sample interaction object within a historical time period, and determine the positive sample recommendation objects and negative sample recommendation objects in the sample recommendation object sequence; the positive sample recommendation object refers to the sample recommendation object that generated the target interaction behavior, and the negative sample recommendation object refers to the sample recommendation object that did not generate the target interaction behavior; Based on the conversion reasons of the positive sample recommendation objects corresponding to the target interaction behavior and the conversion reasons of the negative sample recommendation objects corresponding to the target interaction behavior, the preference conversion reasons of the sample interaction objects when generating the target interaction behavior are determined, and the profile features of the sample interaction objects are generated based on the preference conversion reasons of the sample interaction objects. The step of determining the second object feature of the sample interaction object based on the first object feature of each sample recommendation object in the sample recommendation object sequence within a historical time period includes: The first object features of each sample recommendation object in the sample recommendation object sequence and the profile features of the sample interaction object are concatenated to obtain the second object features of the sample interaction object.

5. The method according to claim 4, characterized in that, Before performing feature representation on the sample recommendation object based on the conversion reason corresponding to the target interaction behavior to obtain the first object feature of the sample recommendation object, the method further includes: Based on the reasons for preference conversion stated by each sample interaction object, determine the set of reasons for preference conversion; For each sample recommendation object corresponding to the target interaction behavior, the conversion reason is adjusted based on the set of preference conversion reasons to obtain the adjusted conversion reason for each sample recommendation object; the adjusted conversion reason is a subset of the set of preference conversion reasons. The adjusted conversion reason corresponding to each of the sample recommendation objects is used as the conversion reason of the sample recommendation object corresponding to the target interaction behavior. The step of performing feature representation on the sample recommendation object based on the conversion reason of the sample recommendation object corresponding to the target interaction behavior is executed to obtain the first object feature of the sample recommendation object.

6. The method according to claim 4, characterized in that, The step of determining the preference conversion reason of the sample interaction object when generating the target interaction behavior based on the conversion reason of the positive sample recommendation object corresponding to the target interaction behavior and the conversion reason of the negative sample recommendation object corresponding to the target interaction behavior, and generating the profile features of the sample interaction object based on the preference conversion reason of the sample interaction object, includes: Obtain a second preset thought chain prompt text; the second preset thought chain prompt text is used to prompt the analysis of the preferences of the sample interaction object when generating the target interaction behavior in multiple steps; The conversion reasons of the positive sample recommendation object corresponding to the target interaction behavior and the conversion reasons of the negative sample recommendation object corresponding to the target interaction behavior are input into the large language model. The large language model performs preference analysis of the sample interaction object based on the second preset thinking chain prompt text to obtain the profile features of the sample interaction object. The multiple steps include, in sequence: identifying positive sample recommendation objects that generate the target interaction behavior and negative sample recommendation objects that do not generate the target interaction behavior; comparing the conversion reasons of the positive sample recommendation objects with the conversion reasons of the negative sample recommendation objects; analyzing the preference conversion reasons of the sample interaction objects when generating the target interaction behavior based on the differences; and predicting the type and degree of preference of the recommended objects preferred by the sample interaction objects based on the preference conversion reasons.

7. The method according to claim 1, characterized in that, The reasoning process for generating the target interactive behavior for the sample recommendation object includes reasoning from multiple preset dimensions associated with the target interactive behavior.

8. The method according to claim 1, characterized in that, The process of obtaining the object description information of the sample recommendation object includes: Obtain information on multiple modalities of the sample recommendation object; Based on one or more of the information from the multiple modalities, determine the object description information of the sample recommendation object.

9. An object feature determination device, characterized in that, include: The object description acquisition unit is configured to acquire the object description information of any sample recommendation object in the sample recommendation object set. The conversion analysis unit is configured to input the object description information into a large language model, and then perform conversion analysis of the target interaction behavior based on a first preset thought chain prompt text through the large language model to obtain the conversion reason for the sample recommended object corresponding to the target interaction behavior; the first preset thought chain prompt text is used to prompt the reasoning process for generating the target interaction behavior for the sample recommended object; The reason for conversion includes the reasoning result of at least one dimension in the reasoning process, and the reasoning result of each dimension represents the performance of the sample recommendation object in that dimension in generating the target interactive behavior; The first feature representation unit is configured to perform feature representation on the sample recommendation object based on the conversion reason corresponding to the target interaction behavior of the sample recommendation object, so as to obtain the first object feature of the sample recommendation object; The second feature representation unit is configured to perform a process of determining the second object feature of the sample interaction object based on the first object feature of each sample recommendation object in the sample recommendation object sequence within a historical time period; wherein the first object feature of the sample recommendation object and the second object feature of the sample interaction object are used to train a recommendation probability prediction model, and the recommendation probability prediction model is used to predict the probability of generating the target interaction behavior.

10. An electronic device, characterized in that, include: processor; Memory used to store the processor's executable instructions; The processor is configured to execute the instructions to implement the object feature determination method as described in any one of claims 1 to 8.

11. A computer-readable storage medium, characterized in that, When the instructions in the computer-readable storage medium are executed by the processor of the electronic device, the electronic device is able to perform the object feature determination method as described in any one of claims 1 to 8.

12. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the object feature determination method as described in any one of claims 1 to 8.

Images