Label prediction method, label acquisition method and related product

Abstract

The invention discloses a label prediction method, a label acquisition method and a related product. The method comprises the steps that target features and a target model of a target object are obtained, the target model is a model used for predicting an interest label of the object, the interest label is a label of a material interested by the object, the target model is obtained through training of training data, and the training data comprises training features and training labels of a training object; the training label represents a label of a material interested by the training object; the training label is obtained by utilizing language model prediction, and a prediction reason of the training label obtained by language model prediction meets a preset condition; the target features serve as input of a target model, so that the target model predicts potential interest tags of the target object based on the target features, and the potential interest tags are tags of materials interested by the target object.

Description

Technical Field

[0001] The present application relates to the field of deep learning technology, and in particular to a label prediction method, a label acquisition method and related products. Background Art

[0002] In the recommendation scenario, by determining the label of the material that the subject is interested in, the material that the subject is interested in can be further recommended to the subject based on the label. Therefore, how to determine the label of the material that the subject is interested in is of great significance. Summary of the invention

[0003] The present application provides a label prediction method, a label acquisition method and related products, wherein the related products include a label prediction device, a label acquisition device, an electronic device, a computer-readable storage medium and a computer program product.

[0004] In a first aspect, a label prediction method is provided, the method comprising: obtaining a target feature and a target model of a target object, the target model being a model for predicting an interest label of the object, the interest label being a label of a material of interest to the object, the target model being obtained by training with training data, the training data comprising training features and training labels of the training object, the training labels representing labels of materials of interest to the training object; the training labels being obtained by prediction using a language model, and the prediction reason for the training labels predicted by the language model meeting a preset condition;

[0005] The target feature is used as an input of the target model, so that the target model predicts potential interest tags of the target object based on the target feature, and the potential interest tags are tags of materials that the target object is interested in.

[0006] In combination with any embodiment of the present application, after taking the target feature as the input of the target model so that the target model predicts the potential interest tag of the target object based on the target feature, the method further includes:

[0007] Based on the similarity between the n preset tags and the potential interest tags, determining a target tag from the n preset tags;

[0008] The target tag is used to determine the recommended material for the target object.

[0009] In combination with any embodiment of the present application, the target feature is used as the input of the target model so that the target model predicts the potential interest tag of the target object based on the target feature, and the potential interest tag is a tag of the material that the target object is interested in, including:

[0010] Acquire an existing interested tag of the target object, where the existing interested tag is a tag of a material that the target object is interested in;

[0011] Using the target model to predict tags of materials that the target object is interested in based on the target features, and obtaining candidate tags of interest;

[0012] When the candidate interest tag is related to the existing interest tag and the candidate interest tag is different from the existing interest tag, the candidate interest tag is used as the potential interest tag.

[0013] In combination with any embodiment of the present application, the n preset tags include i parent tags and (ni) child tags of the i parent tags, and the parent tag corresponds to one or more of the child tags;

[0014] The determining a target tag from the n preset tags based on the similarity between the n preset tags and the potential interest tag comprises:

[0015] Determine the tag with the highest similarity to the potential interest tag from the i parent tags to obtain a candidate tag;

[0016] When the similarity between the candidate tag and the potential interest tag is greater than or equal to a similarity threshold, taking the candidate tag as the target tag;

[0017] In the case where the similarity between the candidate tag and the potential interest tag is less than the similarity threshold, determining the subtag of the candidate tag from the (ni) subtags to obtain y candidate tags, where y is less than (ni);

[0018] Determine the tag with the highest similarity to the potential interest tag from the y candidate tags to obtain the target tag.

[0019] In combination with any embodiment of the present application, determining the tag with the highest similarity to the potential interest tag from the n preset tags to obtain the target tag includes:

[0020] Acquire s reference tags of a reference object, wherein the reference tags are tags of materials of interest to the reference object, the reference tags are tags among the n preset tags, and s is a positive integer less than n;

[0021] Extracting an embedding of a reference feature of the reference object to obtain a reference feature embedding;

[0022] Extracting the embedding of the target feature of the target object to obtain a target feature embedding;

[0023] When the reference feature embedding matches the target feature embedding, a tag with the highest similarity to the potential interest tag is determined from the s reference tags to obtain the target tag.

[0024] In a second aspect, a tag acquisition method is provided, the method comprising:

[0025] Acquire training features and a language model of a training object; the language model is a model for predicting interest tags of the object, and the interest tags are tags of materials that the object is interested in;

[0026] Generate a target prompt word according to the training features, wherein the target prompt word is used to guide the language model to predict the label of the material of interest to the training subject according to the training features, and to generate a reason for the predicted label;

[0027] Under the guidance of the target prompt word, using the language model to predict the label of the material that the training subject is interested in based on the training features to obtain a predicted label;

[0028] Under the guidance of the target prompt word, using the language model to generate a prediction reason for obtaining the predicted label;

[0029] In the case where the prediction reason meets a preset condition, the prediction label is used as a training label, and the training label is a label of the material that the training object is interested in.

[0030] In combination with any implementation manner of the present application, the preset condition includes a confidence threshold;

[0031] When the prediction reason satisfies a preset condition, determining the prediction label as a training label includes:

[0032] When the confidence of the prediction reason is greater than the confidence threshold, the prediction label is used as the training label.

[0033] In conjunction with any embodiment of the present application, when the confidence of the prediction reason is greater than the confidence threshold, before using the prediction label as the training label, the method further includes:

[0034] In a case where the prediction reason matches the training feature, it is determined that the confidence of the prediction reason is greater than the confidence threshold.

[0035] In combination with any embodiment of the present application, the target prompt word also includes a real interest tag of the training object;

[0036] In the case where the confidence of the prediction reason is greater than the confidence threshold, before using the prediction label as the training label, the method further includes:

[0037] When the real interest tag is related to the predicted reason and the real interest tag is different from the predicted tag, it is determined that the confidence of the predicted reason is greater than the confidence threshold.

[0038] In a third aspect, a label prediction method is provided, the method comprising:

[0039] Obtaining target features and a prediction model of a target object, wherein the prediction model is a model for predicting an interest tag of the object, and the interest tag is a tag of a material of interest to the object;

[0040] Generate a prediction prompt word according to the target feature, wherein the prediction prompt word is used to guide the prediction model to predict the label of the material of interest to the target object according to the target feature, and generate a reason for the predicted label;

[0041] Under the guidance of the prediction prompt word, the prediction model is used to predict the label of the material that the target object is interested in based on the target feature to obtain a first label;

[0042] Under the guidance of the prediction prompt word, using the prediction model to generate a first reason for predicting the first label;

[0043] In a case where it is determined based on the first reason that the confidence of the first tag is greater than or equal to a tag threshold, a tag of the material of interest to the target object is determined based on the first tag.

[0044] In combination with any embodiment of the present application, when it is determined based on the first reason that the confidence of the first tag is greater than or equal to the tag threshold, before determining the tag of the material of interest to the target object based on the first tag, the method further includes:

[0045] When the confidence of the first reason is greater than a confidence threshold, it is determined that the confidence of the first label is greater than or equal to the label threshold.

[0046] In combination with any embodiment of the present application, when the confidence of the first reason is greater than the confidence threshold, before determining that the confidence of the first label is greater than or equal to the label threshold, the method further includes:

[0047] In a case where the predicted reason matches the target feature, it is determined that the confidence of the first reason is greater than the confidence threshold.

[0048] In a fourth aspect, a label prediction device is provided, the label prediction device comprising:

[0049] An acquisition unit is used to acquire a target feature and a target model of a target object, wherein the target model is a model for predicting an interest tag of the object, wherein the interest tag is a tag of a material of interest to the object, and wherein the target model is obtained by training with training data, wherein the training data includes training features and training tags of the training object, and wherein the training tags represent tags of materials of interest to the training object; wherein the training tags are obtained by prediction using a language model, and a prediction reason for the training tags predicted by the language model satisfies a preset condition;

[0050] A prediction unit is used to use the target feature as an input of the target model so that the target model predicts a potential interest tag of the target object based on the target feature, wherein the potential interest tag is a tag of a material that the target object is interested in.

[0051] In combination with any embodiment of the present application, the label prediction device further includes a mapping unit, which is used to:

[0052] Based on the similarity between the n preset tags and the potential interest tags, determining a target tag from the n preset tags;

[0053] The target tag is used to determine the recommended material for the target object.

[0054] In conjunction with any embodiment of the present application, the prediction unit is further used to:

[0055] Acquire an existing interested tag of the target object, where the existing interested tag is a tag of a material that the target object is interested in;

[0056] Using the target model to predict tags of materials that the target object is interested in based on the target features, and obtaining candidate tags of interest;

[0057] When the candidate interest tag is related to the existing interest tag and the candidate interest tag is different from the existing interest tag, the candidate interest tag is used as the potential interest tag.

[0058] In combination with any embodiment of the present application, the n preset tags include i parent tags and (ni) child tags of the i parent tags, and the parent tag corresponds to one or more of the child tags;

[0059] The above mapping unit is also used for:

[0060] Determine the tag with the highest similarity to the potential interest tag from the i parent tags to obtain a candidate tag;

[0061] When the similarity between the candidate tag and the potential interest tag is greater than or equal to a similarity threshold, taking the candidate tag as the target tag;

[0062] In the case where the similarity between the candidate tag and the potential interest tag is less than the similarity threshold, determining the subtag of the candidate tag from the (ni) subtags to obtain y candidate tags, where y is less than (ni);

[0063] Determine the tag with the highest similarity to the potential interest tag from the y candidate tags to obtain the target tag.

[0064] In combination with any embodiment of the present application, the mapping unit is further used for:

[0065] Acquire s reference tags of a reference object, wherein the reference tags are tags of materials of interest to the reference object, the reference tags are tags among the n preset tags, and s is a positive integer less than n;

[0066] Extracting an embedding of a reference feature of the reference object to obtain a reference feature embedding;

[0067] Extracting the embedding of the target feature of the target object to obtain a target feature embedding;

[0068] When the reference feature embedding matches the target feature embedding, a tag with the highest similarity to the potential interest tag is determined from the s reference tags to obtain the target tag.

[0069] In a fifth aspect, a label acquisition device is provided, the label acquisition device comprising:

[0070] An acquisition unit, used to acquire training features and a language model of a training object; the language model is a model for predicting interest tags of an object, and the interest tags are tags of materials that the object is interested in;

[0071] A generating unit, configured to generate a target prompt word according to the training feature, wherein the target prompt word is used to guide the language model to predict a label of the material of interest to the training subject according to the training feature, and to generate a reason for the predicted label;

[0072] A prediction unit, used to predict the label of the material of interest to the training subject based on the training features using the language model under the guidance of the target prompt word, to obtain a predicted label;

[0073] The prediction unit is further used to generate a prediction reason for the predicted label using the language model under the guidance of the target prompt word;

[0074] A determination unit is used to use the predicted label as a training label when the prediction reason meets a preset condition, and the training label is a label of the material that the training object is interested in.

[0075] In conjunction with any embodiment of the present application, the preset condition includes a confidence threshold;

[0076] The above-mentioned determination unit is specifically used for:

[0077] When the confidence of the prediction reason is greater than the confidence threshold, the prediction label is used as the training label.

[0078] In combination with any embodiment of the present application, the above-mentioned determination unit is further used to:

[0079] In a case where the prediction reason matches the training feature, it is determined that the confidence of the prediction reason is greater than the confidence threshold.

[0080] In combination with any embodiment of the present application, the target prompt word also includes a real interest tag of the training object;

[0081] The above determination unit is further used for:

[0082] When the real interest tag is related to the predicted reason and the real interest tag is different from the predicted tag, it is determined that the confidence of the predicted reason is greater than the confidence threshold.

[0083] In a sixth aspect, a label prediction device is provided, the label prediction device comprising:

[0084] An acquisition unit, used to acquire target features and a prediction model of a target object, wherein the prediction model is a model for predicting an interest tag of the object, and the interest tag is a tag of a material of interest to the object;

[0085] A generating unit, configured to generate a prediction prompt word according to the target feature, wherein the prediction prompt word is used to guide the prediction model to predict the label of the material of interest to the target object according to the target feature, and to generate a reason for the predicted label;

[0086] A prediction unit, configured to predict a label of the material of interest to the target object based on the target feature using the prediction model under the guidance of the prediction prompt word, to obtain a first label;

[0087] The prediction unit is further configured to generate a first reason for predicting the first label using the prediction model under the guidance of the prediction prompt word;

[0088] A determination unit is configured to determine a label of the material of interest to the target object based on the first label when it is determined based on the first reason that the confidence of the first label is greater than or equal to a label threshold.

[0089] In combination with any embodiment of the present application, the above-mentioned determination unit is further used to:

[0090] When the confidence of the first reason is greater than a confidence threshold, it is determined that the confidence of the first label is greater than or equal to the label threshold.

[0091] In conjunction with any embodiment of the present application, the above-mentioned determination unit is specifically used to:

[0092] In a case where the predicted reason matches the target feature, it is determined that the confidence of the first reason is greater than the confidence threshold.

[0093] In the seventh aspect, an electronic device is provided, comprising: a processor and a memory, the memory being used to store computer program code, the computer program code comprising computer instructions, and when the processor executes the computer instructions, the electronic device executes the above-mentioned first aspect and any embodiment thereof, or executes the above-mentioned second aspect and any embodiment thereof, or executes the above-mentioned third aspect and any embodiment thereof.

[0094] In an eighth aspect, another electronic device is provided, comprising: a processor, a sending device, an input device, an output device and a memory, wherein the memory is used to store computer program code, and the computer program code includes computer instructions. When the processor executes the computer instructions, the electronic device executes the first aspect and any embodiment thereof, or executes the second aspect and any embodiment thereof, or executes the third aspect and any embodiment thereof.

[0095] In the ninth aspect, a computer-readable storage medium is provided, wherein a computer program is stored in the computer-readable storage medium, and the computer program includes program instructions. When the program instructions are executed by a processor, the processor is caused to execute the first aspect and any embodiment thereof, or execute the second aspect and any embodiment thereof, or execute the third aspect and any embodiment thereof.

[0096] In the tenth aspect, a computer program product is provided, which includes a computer program or instructions, and when the computer program or instructions are run on a computer, the computer is caused to execute the above-mentioned first aspect and any embodiment thereof, or execute the above-mentioned second aspect and any embodiment thereof, or execute the above-mentioned third aspect and any embodiment thereof.

[0097] It is to be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present application.

[0098] In the present application, the training data includes training features and training labels of the training object, wherein the training label represents the label of the material of interest to the training object, and the training label is predicted by using a language model, and the prediction reason of the training label predicted by the language model meets the preset conditions, so the accuracy of the training label is high. In this way, when the target model is trained by the training data, the target model can predict the interest label of the material of interest to the object based on the features of the object, and because the accuracy of the training label is high, the accuracy of the interest label predicted by the target model is high. Therefore, after obtaining the target features and the target model of the target object, the label prediction device uses the target features as the input of the target model, so that the target model can predict the potential interest labels of the material of interest to the target object based on the target features, and the accuracy of the potential interest labels can be improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0099] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the background technology, the drawings required for use in the embodiments of the present application or the background technology will be described below.

[0100] The drawings herein are incorporated into the specification and constitute a part of the specification. These drawings illustrate embodiments consistent with the present application and are used together with the specification to illustrate the technical solution of the present application.

[0101] Figure 1 A flowchart of a label prediction method provided in an embodiment of the present application;

[0102] Figure 2 A flowchart of a tag acquisition method provided in an embodiment of the present application;

[0103] Figure 3 A schematic diagram of a label acquisition device provided in an embodiment of the present application performing label acquisition;

[0104] Figure 4 A schematic diagram of a flow chart of another label prediction method provided in an embodiment of the present application;

[0105] Figure 5 A schematic diagram of the structure of a label prediction device provided in an embodiment of the present application;

[0106] Figure 6 A schematic diagram of the structure of a tag acquisition device provided in an embodiment of the present application;

[0107] Figure 7 A schematic diagram of the structure of another label prediction device provided in an embodiment of the present application;

[0108] Figure 8 A schematic diagram of the hardware structure of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0109] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0110] The terms "first", "second", etc. in the specification and claims of this application and the above-mentioned drawings are used to distinguish different objects, rather than to describe a specific order. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device that includes a series of steps or units is not limited to the listed steps or units, but optionally includes steps or units that are not listed, or optionally includes other steps or units inherent to these processes, methods, products or devices.

[0111] Reference to "embodiments" herein means that a particular feature, structure, or characteristic described in conjunction with the embodiments may be included in at least one embodiment of the present application. The appearance of the phrase in various locations in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment that is mutually exclusive with other embodiments. It is explicitly and implicitly understood by those skilled in the art that the embodiments described herein may be combined with other embodiments.

[0112] The execution subject of the embodiment of the present application is a tag prediction device, wherein the tag prediction device can be any electronic device that can execute the technical solution disclosed in the method embodiment of the present application. Optionally, the tag prediction device can be one of the following: a computer, a server.

[0113] It should be understood that the method embodiment of the present application can also be implemented by a processor executing a computer program code. The following describes the embodiment of the present application in conjunction with the drawings in the embodiment of the present application. Figure 1 , Figure 1 A flowchart of a label prediction method provided in an embodiment of the present application.

[0114] 101. Obtain target features and target models of the target object.

[0115] In an embodiment of the present application, the target model is a model for predicting the interest tag of an object. The target model is a model trained using training data. The training data includes training features of the training object and training tags of the training object. The training tags are interest tags predicted using a language model. The training tags are related to the training features of the training object. The interest tags are tags of materials that the object is interested in. The reason for predicting the training tags using the language model satisfies a preset condition. In one possible implementation, the preset condition includes m necessary features. When any necessary feature exists in the prediction reason, it is determined that the preset reason satisfies the preset condition. In another possible implementation, the preset condition includes a confidence threshold. When the confidence of the prediction reason is higher than the confidence threshold, it is determined that the preset reason satisfies the preset condition.

[0116] In the embodiment of the present application, both the target model and the language model have the ability of natural language processing. In a possible implementation, the natural language processing ability of the target model and the language model is obtained by training with text training data, wherein the text training data includes one or more of masked text and non-masked text. Masked text is a text including a mask. When the text training data includes masked text, the language model is trained using the text training data, so that the language model can predict the content blocked by the mask in the masked text to obtain a first prediction result, and then the loss of the language model is determined based on the difference between the first prediction result and the ground truth (GT) of the mask. The parameters of the language model are updated based on the loss until the loss of the language model converges, and the training of the language model can be completed. Non-masked text is a text that does not include a mask. When the text training data does not include non-masked text, the language model is trained using text training data, so that the language model can perform natural language processing tasks using non-masked text, thereby improving the natural language processing ability of the language model, wherein the natural language processing tasks include: summarizing the summary of the text, determining the order of sentences, and determining the answer to the question based on the text.

[0117] Optionally, the target model and language model are large language models (LLM), where LLM refers to a deep learning model trained with a large amount of text data that can generate natural language text or understand the meaning of language text. Large language models can handle a variety of natural language tasks, such as text classification, question answering, dialogue, generating text summaries, and generating code.

[0118] In the embodiment of the present application, the target feature represents the basic information of the target object, and the target feature is the unique identifier of the target object. The training feature represents the basic information of the training object, and the training feature is the unique identifier of the training object.

[0119] In a possible implementation, the target object is the target user who consumes materials on the material consumption platform, the training object is the training user who consumes materials on the material consumption platform, the target characteristics of the target user represent the target user's age, gender, city, etc., and the training characteristics of the training user represent the training user's age, gender, city, etc. When it is necessary to collect the target characteristics of the target user and the training characteristics of the training user, use obvious signs / information to inform the personal information processing rules, and obtain personal authorization through pop-up information or by the individual uploading their personal information. Among them, personal information processing may include information such as the personal information processor, the purpose of personal information processing, the processing method, and the type of personal information processed.

[0120] In one implementation of obtaining target features and target models, the tag prediction receives the target features and target models input by an input component. The input component includes at least one of the following: a keyboard, a mouse, a touch screen, a touch pad, and an audio input device.

[0121] In another implementation of obtaining the target features and the target model, the tag prediction receives the target features and the target model sent by the terminal. The terminal includes at least one of the following: a mobile phone, a computer, a tablet computer, and a server.

[0122] It should be understood that in the embodiments of the present application, the steps of label prediction to obtain target features and target models can be performed separately or simultaneously, and the present application does not limit this.

[0123] 102. Using the target features as input to the target model, so that the target model predicts potential interest tags of the target object based on the target features.

[0124] In the embodiment of the present application, the potential interest tag is the interest tag of the target object. When the tag prediction device determines the target feature of the target object, it uses the target feature as the input of the target model and uses the target model to predict the interest tag related to the target feature, that is, the potential interest tag.

[0125] Generally speaking, when a tag is predicted, you need to determine the target object's interest in the tag. If the target object is interested in the tag, use the tag as the target object's interest tag. For example, when the predicted tag is game, you need to determine the target object's interest in the game. If the target object is interested in the game, use the game tag as the target object's interest tag.

[0126] In the embodiment of the present application, the tag prediction device uses the target model to predict the target object's interest tags as potential interest tags. Since the target model is trained by training data, and the training data includes the training features of the training object and the training tags of interest to the training object, the potential interest tags predicted by the target model are more likely to be tags of interest to the target object.

[0127] In one possible implementation, the tag prediction device generates a feature prompt based on the target feature of the target object, and the feature prompt is used to guide the tag prediction device to use the target model to predict interest tags related to the target feature as potential interest tags.

[0128] In an embodiment of the present application, the training data includes training features and training labels of the training object, wherein the training label represents the label of the material of interest to the training object, and the training label is predicted by using a language model, and the prediction reason of the training label predicted by the language model satisfies a preset condition, so the accuracy of the training label is high. In this way, when the target model is trained by the training data, the target model can predict the interest label of the material of interest to the object based on the features of the object, and because the accuracy of the training label is high, the accuracy of the interest label predicted by the target model is high. Therefore, after obtaining the target features and the target model of the target object, the label prediction device uses the target features as the input of the target model, so that the target model can predict the potential interest labels of the material of interest to the target object based on the target features, and the accuracy of the potential interest labels can be improved.

[0129] As an optional implementation manner, the label prediction device performs the following steps after executing step 102:

[0130] 2001. Based on the similarity between the n preset tags and the potential interest tags, determine a target tag from the n preset tags.

[0131] In an embodiment of the present application, n preset tags include tags of materials to be recommended, wherein the materials to be recommended include materials to be recommended to the target object, wherein the materials to be recommended include: one or more of documents, images, and videos. For example, n preset tags include: games, beauty, and sports. The materials to be recommended include documents to be recommended to the target object, and the documents recommended to the target object can be determined from the documents. The tags of any document include one or more of games, beauty, and sports.

[0132] Since the n predicted tags include the tags of the materials to be recommended to the target object, before determining the materials recommended to the target object from the materials to be recommended to the target object, it is necessary to map the potential interest tags to one of the n preset tags, that is, it is necessary to determine the tags that match the potential interest tags of the target object from the n predicted tags, and then the materials recommended to the target object can be determined based on the tags that match the potential interest tags of the target object.

[0133] The target tag obtained by executing step 2001 is the tag determined to match the potential interest tag of the target object among the n predicted tags. Since the potential interest tag is the tag of the material that the target object is interested in, the target tag is the tag of the material that the target object is interested in. In a possible implementation, in order to determine the target tag from the n preset tags, the tag prediction device calculates the similarity between the n preset tags and the potential interest tag to obtain n reference similarities, determines the maximum similarity from the n reference similarities, and then determines the tag with the maximum similarity as the target tag.

[0134] In another possible implementation, when the number of preset tags is large, the tag prediction device takes a long time to calculate the similarity of n preset tags. Therefore, the embedding of the target feature, that is, the target feature embedding, can be used to retrieve the n preset tags, determine the x embedded tags that match the target feature embedding from the n preset tags, and then determine the tag with the highest similarity to the potential interest tag from the x embedded tags as the target tag. The above method reduces the number of times the tag prediction device calculates the similarity from n times to x times, thereby speeding up the speed at which the tag prediction device obtains the target tag. Optionally, the tag prediction device determines a reference feature embedding that matches the target feature embedding, wherein the reference feature embedding is an embedding of a reference feature of a reference object. When the target feature embedding matches the reference feature embedding, the s reference tags of interest to the reference object are used as the above x embedded tags, and the reference tags are tags in the preset tags. When the x embedded tags are determined to be s reference tags, the tag with the highest similarity to the above potential interest tag is determined from the above s reference tags to obtain the above target tag.

[0135] 2002. Determine the recommended materials for the target object using the target tags.

[0136] By executing step 2002, the label prediction device can determine that the materials recommended to the target object include materials matching the target label, which are materials of interest to the target object. In addition, materials matching the target label (i.e., materials of interest to the target object) can be recommended to the target object. In one possible implementation (the material is a document), the label prediction device determines that the documents matching the target label are the documents recommended to the target object. For example, the materials to be recommended to the target object include document A, document B, and document C, wherein the label of document A is games, the label of document B is beauty, and the label of document C is sports. When the target label is sports, document C is a document matching the target label, and therefore document C is determined to be the document recommended to the target object.

[0137] In this embodiment, the tag prediction device determines a target tag that matches the potential interest tag from the n preset tags based on the similarity between the n preset tags and the potential interest tag, wherein the n predicted tags include the tag of the material to be recommended to the target object. Then, the target tag can be used to determine the material that the target object is interested in from the materials to be recommended to the target object as the material recommended to the target object.

[0138] As an optional implementation manner, the label prediction device performs the following steps during the process of executing step 102:

[0139] 3001. Obtain the existing interested tags of the target object.

[0140] In the embodiment of the present application, the existing interest tag is an existing interest tag of the target object in the material consumption platform. Optionally, the existing interest tag is a tag of the material of interest to the target object determined by the behavior data of the target object in the material consumption platform. For example, when the behavior data of the target object in the material consumption platform includes more than z pieces of data of browsing materials related to sports, sports is used as the tag of the material of interest to the target object, that is, the existing interest tag.

[0141] 3002. Using the target model and based on the target features, predict the labels of the materials that the target object is interested in, and obtain candidate interest labels.

[0142] In the embodiment of the present application, the candidate interest tags are tags predicted by the tag prediction device that the target object is interested in. Specifically, the tag prediction device uses the target feature as the input of the target model, and then uses the target model to generate the candidate interest tags based on the target feature.

[0143] 3003. When the candidate interest tag is related to the existing interest tag and the candidate interest tag is different from the existing interest tag, the candidate interest tag is used as the potential interest tag.

[0144] In an embodiment of the present application, when a candidate interest tag is related to an existing interest tag, and the candidate interest tag is different from the existing interest tag, the candidate interest tag is a tag of a material predicted based on the existing interest tag, and the candidate interest tag is a different tag from the existing interest tag, indicating that the candidate interest tag is an extension of the existing interest tag. Since the existing interest tag is a tag that the target object is interested in, and the candidate interest tag is an extension of the existing interest tag, the probability that the candidate interest tag is a tag of a material that the target object is interested in is high, and using the candidate interest tag as a potential interest tag can make the potential interest tag also an extension of the existing interest tag. For example, the existing interest tag is beauty products, and the candidate interest tag is beauty skills. Both the existing interest tag and the candidate interest tag are tags related to beauty, so the existing interest tag is related to the candidate interest tag, and the existing interest tag is a tag related to the product, and the candidate interest tag is a tag related to the skills, so the existing interest tag and the candidate interest tag are related and different, so the candidate interest tag is an extension of the existing interest tag. Using the candidate interest tag as a potential interest tag can make the potential interest tag also an extension of the existing interest tag.

[0145] In a possible implementation, the degree of relevance between the candidate interest tag and the existing interest tag is determined by using the degree of match between the candidate interest tag and the existing interest tag. When the degree of match between the candidate interest tag and the existing interest tag is greater than the first matching threshold, it means that the degree of match between the candidate interest tag and the existing interest tag is high, that is, the candidate interest tag is related to the existing interest tag, and when the degree of match between the candidate interest tag and the existing interest tag is less than the second threshold, it means that the candidate interest tag is different from the existing interest tag, wherein the second threshold is greater than the first matching threshold. Therefore, when the degree of match between the candidate interest tag and the existing interest tag is greater than the first matching threshold and less than the second threshold, the candidate interest tag is related to and different from the existing interest tag, that is, the candidate interest tag is an extension of the existing interest tag. In the case where the candidate interest tag is an extension of the existing interest tag, since the existing interest tag is a label of the material of interest to the target object, the probability that the candidate interest tag is a label of the material of interest to the target object is high, and the candidate interest tag is used as a potential interest tag.

[0146] In another possible implementation, the degree of relevance between the candidate interest tag and the existing interest tag is determined by using the similarity between the semantics of the candidate interest tag and the semantics of the existing interest tag. When the similarity between the semantics of the candidate interest tag and the semantics of the existing interest tag is greater than the first similarity threshold, it means that the candidate interest tag has a high similarity with the existing interest tag, that is, the candidate interest tag is related to the existing interest tag, and when the similarity between the semantics of the candidate interest tag and the semantics of the existing interest tag is less than the second similarity threshold, it means that the candidate interest tag is different from the existing interest tag, wherein the first similarity threshold is less than the second similarity threshold. Therefore, when the similarity between the semantics of the candidate interest tag and the semantics of the existing interest tag is greater than the first similarity threshold and less than the second similarity threshold, the candidate interest tag is related to and different from the existing interest tag, that is, the candidate interest tag is an extension of the existing interest tag. In the case where the candidate interest tag is an extension of the existing interest tag, since the existing interest tag is a tag of the material of interest to the target object, the probability that the candidate interest tag is a tag of the material of interest to the target object is high, and the candidate interest tag is used as a potential interest tag.

[0147] In another possible implementation, a classification model is used to classify the keywords of the candidate interest tags and the keywords of the existing interest tags to determine the degree of relevance between the candidate interest tags and the existing interest tags. The classification model will divide different keywords into different areas. When the area to which the keywords of the candidate interest tags belong is different from the area to which the keywords of the existing interest tags belong, it means that the candidate interest tags are different from the existing interest tags. When there is an intersection between the above two areas, it means that the candidate interest tags are related to the existing interest tags. Therefore, when the above two areas are different and there is an intersection, the candidate interest tags are related and different from the existing interest tags, that is, the candidate interest tags are extensions of the existing interest tags. In the case where the candidate interest tags are extensions of the existing interest tags, since the existing interest tags are tags of materials of interest to the target object, the probability that the candidate interest tags are tags of materials of interest to the target object is high, and the candidate interest tags are used as potential interest tags.

[0148] In this embodiment, the tag prediction device generates a candidate interest tag related to the target feature. When the candidate interest tag is related to the existing interest tag of the target object, and the candidate interest tag is different from the existing interest tag, it means that the candidate interest tag is an extension of the existing interest tag. At this time, the candidate interest tag is used as a potential interest tag, which can make the potential interest tag an extension of the existing interest tag. Since the existing interest tag is a tag that the target object is interested in, when the candidate interest tag is an extension of the existing interest tag, the probability that the candidate interest tag is a tag that the target object is interested in is high, so the candidate interest tag is used as a potential interest tag, and the probability that the potential interest tag is a tag that the target object is interested in is high.

[0149] As an optional implementation, the n preset tags include i parent tags and (ni) child tags of the i parent tags, and the parent tag corresponds to one or more child tags. The tag prediction device performs the following steps during the execution of step 2001:

[0150] 4001. Determine the tag with the highest similarity to the potential interest tag from the i parent tags to obtain a candidate tag.

[0151] In the embodiment of the present application, the parent tag is i tags among n preset tags, and the child tag is (ni) tags among n preset tags except the parent tag. Any parent tag corresponds to one or more child tags. The similarity between any child tag and the corresponding parent tag is higher than the similarity between any child tag and the non-corresponding parent tag. For example, the parent tags include electronic products and beauty products. The child tags include smartphones. Smartphones correspond to electronic products. The similarity between smartphones and electronic products is higher than the similarity between smartphones and beauty products.

[0152] In an implementation method of obtaining a parent tag and a child tag, the preset tag includes the level of the tag, and the tag is determined to be a parent tag or a child tag by the level of the preset tag. For example, n preset tags include tag 1 and tag 2. The level of tag 1 is a, and the level of tag 2 is a1. The tag prediction device determines tag 1 with a level of a as a parent tag, and determines tag 2 with a level of a1 as a child tag. And there is a corresponding relationship between tag 2 and tag 1.

[0153] In another implementation method of obtaining parent tags and child tags, when the tag prediction device obtains n preset tags, it clusters the n preset tags to obtain i clusters, and any two preset tags in any cluster are related tags. Any preset tags in any two clusters are unrelated tags. The preset tag closest to the centroid of the cluster is used as the parent tag of the cluster, and the tags in the cluster other than the preset tag closest to the centroid of the cluster are used as child tags, and the parent tag corresponds to the child tag.

[0154] In an embodiment of the present application, a tag prediction device uses the similarity between a preset tag and a potential tag of interest to determine a tag of interest to a target object. The greater the similarity between a tag and a potential tag of interest, the higher the probability that the target object is interested in the tag. When determining i parent tags and (ni) child tags, the tag prediction device calculates the similarity between i parent tags and potential tags of interest to obtain i candidate similarities. The maximum candidate similarity is determined from the i candidate similarities, and the parent tag that determines the maximum candidate similarity is used as a candidate tag. The candidate tag corresponds to one or more child tags.

[0155] 4002. When the similarity between the candidate tag and the potential interest tag is greater than or equal to a similarity threshold, the candidate tag is used as the target tag.

[0156] In the embodiment of the present application, after determining the candidate tag, the tag prediction device determines the similarity between the candidate tag and the potential interest tag. The probability that the candidate tag is a tag of interest to the target object is determined by the similarity between the candidate tag and the potential interest tag. When the above similarity is greater than or equal to the similarity threshold, the probability that the candidate tag is a tag of interest to the target object is high, and the tag prediction device uses the candidate tag as the target tag.

[0157] 4003. When the similarity between the candidate tag and the potential interest tag is less than the similarity threshold, the subtag of the candidate tag is determined from the (ni) subtags to obtain y candidate tags.

[0158] In the embodiment of the present application, when the similarity between the candidate tag and the potential interest tag is less than the similarity threshold, the probability that the candidate tag is the tag that the target object is interested in is low. In this case, the subtag corresponding to the candidate tag is determined to obtain y candidate tags.

[0159] 4004. Determine the tag with the highest similarity to the potential interest tag from the y candidate tags to obtain the target tag.

[0160] Since the candidate tag is the tag with the highest similarity to the potential interest tag among the i parent tags, there is a high probability that the maximum similarity among the y candidate tags to the potential interest tag is the maximum similarity among the n preset tags to the potential interest tag, and there is a high probability that the tag with the greatest similarity to the potential interest tag among the y candidate tags is the tag that the target object is interested in among the n preset tags. Specifically, y sub-tags corresponding to the candidate tag are determined, and the similarities between the y sub-tags and the potential interest tag are calculated to obtain y cluster similarities, and the maximum similarity is determined from the y cluster similarities, and then the sub-tag with the maximum similarity is determined as the target tag.

[0161] In the embodiment of the present application, the similarity between the candidate tag and the potential interest tag is the tag with the highest similarity to the potential interest tag among the i parent tags, indicating that the candidate tag is the tag that best matches the potential interest tag among the i parent tags, but because the similarity to the potential interest tag is less than the similarity threshold, it indicates that the probability that the candidate tag is a tag of interest to the target object is low, so the tag of interest to the target object is determined from the y sub-tags corresponding to the candidate tag. Because the above y sub-tags are tags related to the candidate tag, the similarity between the above y sub-tags and the candidate tag is high. In the case where the similarity between the candidate tag and the potential interest tag is the tag with the highest similarity to the potential interest tag among the i parent tags, and the similarity between the candidate tag and the potential interest tag is less than the similarity threshold, the tag with the highest similarity to the potential interest tag is determined among the tags related to the candidate tag, that is, the y sub-tags, which is the tag with the highest similarity to the potential interest tag among the n preset tags. Since the potential interest tag is a tag of interest to the target object, the tag with the highest similarity to the potential interest tag can also be considered as a tag of interest to the target object. In summary, the tag with the highest similarity to the potential interest tag among the y sub-tags is used as the target object's interest tag, that is, the target tag. For example, the parent tag includes tag 1 and tag 3. The sub-tags include tag 2 and tag 4. Tag 2 and tag 4 correspond to tag 1. The similarity between the potential interest tag and tag 1 is 0.6, and the similarity between the potential interest tag and tag 3 is 0.5. Tag 1 is a candidate tag. The similarity threshold is 0.7, and the similarity between the potential interest tag and tag 1 is less than the similarity threshold. Therefore, the similarity between tag 2 and tag 4 and the potential interest tag is determined. The similarity between the potential interest tag and tag 2 is 0.61, and the similarity between the potential interest tag and tag 4 is 0.59. Tag 2 with the maximum similarity of 0.61 is determined as the target tag.

[0162] In this embodiment, the tag prediction device determines n preset tags as i parent tags and (ni) child tags, wherein any parent tag corresponds to one or more child tags, and any child tag has a higher degree of similarity with the corresponding parent tag than with the non-corresponding parent tag. Determine i similarities between the potential interest tag and i parent tags, and use the parent tag with the maximum similarity among the above i similarities as the candidate tag. Since the potential interest tag is a tag of interest to the target object, when the similarity between the candidate tag and the potential interest tag is greater than or equal to the similarity threshold, the probability that the candidate tag is a tag of interest to the target object is high, and the candidate tag is used as the target tag. When the similarity between the candidate tag and the potential interest tag is less than the similarity threshold, it means that the probability that the candidate tag is a tag of interest to the target object is low, but since the candidate tag is the tag most similar to the potential interest tag among the i parent tags, the tag with the highest similarity to the potential interest tag is determined from the child tags corresponding to the candidate tag, which should be the tag with the highest degree of similarity to the potential interest tag among the n preset tags. Since the potential interest tag is a tag of interest to the target object, the tag with the highest degree of similarity to the potential interest tag can also be considered as the tag of interest to the target object. Compared with the method of calculating the similarity between n preset tags and potential interest tags, the above method can reduce the number of times the tag prediction device calculates the similarity, thereby speeding up the speed at which the tag prediction device obtains the target tag.

[0163] See also Figure 2 , Figure 2 A flowchart of a tag acquisition method provided in an embodiment of the present application. The execution subject of the tag acquisition method is a tag acquisition device, wherein the tag acquisition device can be any electronic device that can execute the technical solution disclosed in the embodiment of the method of the present application. Optionally, the tag acquisition device can be one of the following: a computer, a server.

[0164] 201. Obtain training features and language models of the training object.

[0165] In the embodiment of the present application, the language model is a model for predicting the interest tags of the object. Specifically, the language model can obtain tags of materials that are related to the features of the object and are of interest to the object based on the features of the object.

[0166] In the embodiment of the present application, the training feature represents the basic information of the training object, and the training feature is the unique identifier of the training object. The training feature represents the basic information of the training object, and the training feature is the unique identifier of the training object.

[0167] In a possible implementation, the training object is the target user who consumes materials on the material consumption platform, the training object is the training user who consumes materials on the material consumption platform, the training characteristics of the target user represent the age, gender, city, etc. of the target user, and the training characteristics of the training user represent the age, gender, city, etc. of the training user. When it is necessary to collect the training characteristics of the target user and the training characteristics of the training user, the personal information processing rules are notified by obvious signs / information, and personal authorization is obtained through pop-up information or by the individual uploading his or her personal information. Among them, personal information processing may include information such as the personal information processor, the purpose of personal information processing, the processing method, and the type of personal information processed.

[0168] 202. Generate a target prompt word based on the above training features.

[0169] In the embodiment of the present application, the target prompt word is used to guide the language model to predict the label of the material of interest to the training object based on the training features, and to generate the reason for the predicted label, wherein the target prompt word includes the training features.

[0170] In a possible implementation, a first template including a Chain of Thought (COT) is obtained, the COT including the step of predicting the label of the material of interest to the object based on the features of the object, and the step of generating a reason for predicting the label, and the first template includes a first position where the features of the object can be filled in. The training features are filled in the first position of the first template to obtain a target prompt word.

[0171] In another possible implementation, a second template including a few-shot example is obtained, the few-shot includes an example of predicting a label of a material of interest to the object based on the object's features, and an example of generating a reason for the predicted label, and the second template includes a second position in which the object's features can be filled in. The training features are filled in the second position in the second template to obtain a target prompt word.

[0172] 203. Under the guidance of the target prompt word, the language model is used to predict the label of the material that the training subject is interested in based on the training features to obtain a predicted label.

[0173] In the embodiment of the present application, the label acquisition device generates a predicted label related to the training features of the training object using a language model. Specifically, when the label acquisition device generates a target prompt word, the target prompt word is used as the input of the language model. Under the guidance of the target prompt word, the language model predicts the predicted label of the material of interest to the training object based on the training features in the target prompt word.

[0174] 204. Under the guidance of the target prompt word, the language model is used to generate a prediction to obtain a prediction reason for the prediction label.

[0175] In the embodiment of the present application, the prediction reason is the reason why the language model generates the prediction label using the natural language processing capability, and the label acquisition device uses the prediction reason to determine whether the predicted label is the label that the training object is interested in. Specifically, after generating the predicted label, the language model generates the prediction reason for the predicted label under the guidance of the target prompt word.

[0176] Optionally, after the label acquisition device inputs the target prompt word into the language model, the language model, under the guidance of the target prompt word, predicts the predicted label of the material of interest to the training object based on the training features, and generates a prediction reason for the predicted label.

[0177] 205. When the above prediction reason meets the preset conditions, the above prediction label is used as a training label.

[0178] In the embodiment of the present application, the label acquisition device determines whether the prediction reason meets the preset condition when determining the prediction reason. If the prediction reason meets the preset condition, the prediction label is determined as the label of interest to the training object, that is, the training label.

[0179] In a possible implementation, the preset conditions include m necessary features. When any necessary feature exists in the prediction reason, the predicted label is determined as the label of interest to the training object, that is, the training label.

[0180] In another possible implementation, the preset condition includes a confidence threshold, and when the confidence of the prediction reason is higher than the confidence threshold, the predicted label is determined as a label of interest to the training object, that is, a training label.

[0181] In the embodiment of the present application, the label acquisition device generates a target prompt word according to the training feature after acquiring the training feature and language model of the training object. Under the guidance of the target prompt word, the label acquisition device uses the language model to obtain the predicted label based on the training feature and generates a prediction reason for the predicted label. When the prediction reason meets the preset conditions, the predicted label is determined to be the label of interest to the training object, that is, the training label. In this way, the prediction of the label of interest to the training object can be completed without manual labor, thereby reducing labor costs. Moreover, since the training label is obtained when the prediction reason meets the preset conditions, the accuracy of the training label is high.

[0182] The label acquisition device can generate training labels by using the text generation capability of the language model by executing steps 201 to 205, thereby replacing manual labeling of the labels of the materials of interest to the training object. After obtaining the training labels, the target model can be obtained based on the training labels, and then the target model can be executed based on the target model. Figure 1 The tag prediction method predicts potential interest tags of the target object.

[0183] As an optional implementation, the above-mentioned preset condition includes a confidence threshold. The tag acquisition device performs the following steps during the execution of step 205:

[0184] 301. When the confidence of the above-mentioned prediction reason is greater than the above-mentioned confidence threshold, the above-mentioned prediction label is used as the above-mentioned training label.

[0185] In an embodiment of the present application, the label acquisition device determines the confidence of the prediction reason. When the confidence of the prediction reason is greater than a confidence threshold, the predicted label is determined as a label of interest to the training object, that is, a training label.

[0186] In an implementation method for determining the confidence of a prediction reason, a label acquisition device acquires a prior probability, and the prior probability represents a preset probability that the predicted label predicted by the language model is a label of interest to the training object. The probability model is used to determine the likelihood probability that the predicted label is a label of interest to the training object when the prediction reason is a true value. When the prior probability and the likelihood probability are determined, the likelihood probability and the prior probability are combined to obtain a posterior probability, and the posterior probability represents the final probability that the predicted label predicted by the language model is a label of interest to the training object. The confidence of the prediction reason is determined based on the posterior probability, and the confidence is positively correlated with the posterior probability.

[0187] In another implementation of determining the confidence of the prediction reason, when the prediction reason and the training feature match, the prediction reason contains the training feature of the training object, and the confidence of the prediction reason is determined to be greater than a confidence threshold. Optionally, a target text for describing content related to the training feature is determined, and the target text is then retrieved from the prediction reason. When the target text exists in the prediction reason, the prediction reason contains the training feature of the training object, and the confidence of the prediction reason is determined to be greater than the confidence threshold.

[0188] In another implementation method of determining the confidence of the prediction reason, the target prompt word also includes a real interest tag, wherein the real interest tag represents the interest tag that the training object already has in the material consumption platform. After the target prompt word containing the training features and the real interest tag is input into the language model, the predicted tag generated by the language model is related to the training features and the real interest tag. In the case where the real interest tag is related to the prediction reason, and the real interest tag and the predicted tag are different, the confidence of the prediction reason is greater than the confidence threshold, that is, the predicted tag predicted by the language model is an extension of the real interest tag of the training object. For the implementation method of determining that the real interest tag is related to the prediction reason and the real interest tag and the predicted tag are different, please refer to the implementation method of determining that the candidate interest tag is related to the existing interest tag and the candidate interest tag is different from the existing interest tag in step 3003, which will not be repeated here.

[0189] See also Figure 3 , Figure 3A schematic diagram of a label acquisition device provided for the present application for label acquisition. Specifically, the label acquisition device acquires the object information of the training object, and the object information includes the training features and real interest tags of the training object. The training features include: age, gender, city, and birthday. Real interest tags include long-term topics or categories, medium-term topics or categories, and short-term topics or categories. Long-term topics or categories represent that the training object is interested in the topic or category for a time period exceeding the first threshold. Medium-term topics or categories represent that the training object is interested in the topic or category for a time period greater than the second threshold and less than or equal to the first threshold. Short-term topics or categories represent that the training object is interested in the topic or category for a time period less than the second threshold. The training features and real interest tags of the training object are used to generate target prompt words. In addition to the content of the training features and real interest tags, the target prompt words also include: You are a data analyst who is proficient in interest induction. I will provide you with the features and interest tags of the object. You need to analyze the content I provide and give different interest tags to the object. The following are the features and interest tags of the object I provide. Please return different interest tags that meet the above requirements. After generating the target prompt word, the target prompt word is input into the language model. The labels generated by the language model include new book recommendation, funny comics, and beauty tips. The prediction reason for generating the label of new book recommendation is that considering that young people of "25 years old" are generally curious about new things, and the subject is also interested in "reading", so... This prediction reason satisfies the fact that the prediction reason contains the training features of the training object, that is, the age feature, and the label is an extension of the real interest label of the training object, that is, it contains the keywords of the real interest label of the training object, and the keyword is reading. Therefore, the confidence of the prediction reason is greater than the confidence threshold, and the label is a training label. The prediction reason for generating the label of funny comics is that the subject has recently shown interest in "comics", and there are also "funny videos" in the long-term topics or categories, so... This prediction reason satisfies the fact that the label is an extension of the real interest label of the training object, that is, it contains the keywords of the real interest label of the training object, and the keywords are comics and funny videos. Therefore, the confidence of the prediction reason is greater than the confidence threshold, and the label is a training label. The prediction reason for generating the label of beauty tips is that the user's long-term topics or categories, medium-term topics or categories, and short-term topics or categories contain topics or categories related to "beauty and makeup", so... The prediction reason satisfies the requirement that the label is an extension of the real interest label of the training object, that is, it contains the keywords of the real interest label of the training object, and the keyword is beauty and makeup. Therefore, the confidence of the prediction reason is greater than the confidence threshold, and the label is a training label. Optionally, in the case where the training label needs to be used to determine the material recommended to the target object, the label of the material to be recommended is determined, n preset labels are obtained, and the n preset labels are divided into four groups. Specifically, the n preset labels are divided into first-level category labels, second-level category labels, topic labels, and point of interest labels.The tag with the highest similarity to the training tag is determined from the above-mentioned first-level category tags, second-level category tags, topic tags, and point of interest tags, and the tag with the highest similarity is used to determine the material recommended to the target object.

[0190] See also Figure 4 , Figure 4 A flowchart of another tag prediction method provided in an embodiment of the present application. The execution subject of the tag prediction method is a tag prediction device, wherein the tag prediction device can be any electronic device that can execute the technical solution disclosed in the method embodiment of the present application. Optionally, the tag prediction device can be one of the following: a computer, a server.

[0191] 401. Obtain target features and prediction models of the target object.

[0192] In the embodiment of the present application, the prediction model is a model for predicting the interest tags of the object. Specifically, the prediction model has the ability of natural language processing, and can determine the tags of materials that are related to the features of the object and are of interest to the object based on the features of the object.

[0193] For the target features of the target object, please refer to the description of the target features in step 101, which will not be repeated here.

[0194] 402. Generate a prediction prompt word based on the above target features.

[0195] In the embodiment of the present application, the prediction prompt word is used to guide the prediction model to predict the label of the material of interest to the target object based on the target feature, and to generate the reason for predicting the label, wherein the prediction prompt word includes the target feature.

[0196] In a possible implementation, a first template including a COT is obtained, the COT including a step of predicting a label of a material of interest to the object based on a feature of the object, and a step of generating a reason for predicting the label, the first template including a first position where the feature of the object can be filled in. The target feature is filled in the first position of the first template to obtain a predicted prompt word.

[0197] In another possible implementation, a second template including Few-Shot is obtained, the Few-Shot includes an example of predicting a label of a material of interest to the object based on the object's features, and an example of generating a reason for the predicted label, and the second template includes a second position in which the object's features can be filled in. The target features are filled in the second position in the second template to obtain a predicted prompt word.

[0198] 403. Under the guidance of the prediction prompt word, the prediction model is used to predict the label of the material that the target object is interested in based on the target feature to obtain a first label.

[0199] In the embodiment of the present application, the label prediction device generates a first label related to the target feature of the target object using the prediction model. Specifically, when the label prediction device generates a prediction prompt word, the prediction prompt word is used as the input of the prediction model. Under the guidance of the prediction prompt word, the prediction model predicts the first label of the material of interest to the target object based on the target feature in the prediction prompt word.

[0200] 404. Under the guidance of the prediction prompt word, generate a first reason for predicting the first label using the prediction model.

[0201] In the embodiment of the present application, the first reason is the reason why the language model generates the first label using the natural language processing capability, and the label prediction device uses the first reason to determine whether the first label is a label that the training object is interested in. Specifically, after generating the first label, the language model generates the first reason for predicting the first label under the guidance of the prediction prompt word.

[0202] Optionally, after the label prediction device inputs the prediction prompt word into the language model, the language model, under the guidance of the prediction prompt word, predicts the first label of the material of interest to the training object based on the training features, and generates a prediction reason for predicting the first label.

[0203] 405. When it is determined based on the first reason that the confidence of the first tag is greater than or equal to a tag threshold, determine a tag of the material that the target object is interested in based on the first tag.

[0204] In an embodiment of the present application, the label prediction device determines the confidence of the first label based on the first reason when determining the first reason, and determines that the first label is the label of the material of interest to the target object when the confidence of the first label is greater than or equal to the label threshold.

[0205] In a possible implementation, the first reason is input into a language model, and the language model is used to score the first reason predicted by the prediction model to determine the confidence of the first label. When the language model scores the first reason higher than the first confidence threshold, the confidence of the first label is greater than or equal to the label threshold, and the possibility that the first label is the label of the material of interest to the target object is high, so the first label is determined to be the label of the material of interest to the target object.

[0206] In another possible implementation, the first reason is input into the prediction model, and the prediction model is used to determine whether the first reason meets the preset conditions. When the prediction model determines that the first reason meets the preset conditions, it is determined that the confidence of the first label is greater than or equal to the label threshold, and the possibility that the first label is the label of the material of interest to the target object is high, so the first label is determined as the label of the material of interest to the target object. Optionally, the preset conditions include a confidence threshold, and the implementation method of determining the confidence of the first reason refers to the implementation method of determining the confidence of the prediction reason in step 301, which will not be repeated here.

[0207] In an embodiment of the present application, the label prediction device generates a prediction prompt word according to the target feature after obtaining the target feature and the prediction model of the target object, and uses the prediction prompt word as the input of the prediction model. Under the guidance of the prediction prompt word, the prediction model predicts the first label based on the target feature and generates a first reason for predicting the first label, and determines the confidence of the first label based on the first reason. When the confidence of the first label is greater than or equal to the label threshold, the first label is determined to be a label of interest to the target object. In this way, the prediction of the label of the material of interest to the target object can be completed without manual labor, thereby reducing labor costs, and the confidence of the predicted first label is greater than or equal to the label threshold, and the label is more accurate in being the label of the material of interest to the target object.

[0208] Those skilled in the art will appreciate that, in the above method of specific implementation, the order in which the steps are written does not imply a strict execution order and does not constitute any limitation on the implementation process. The specific execution order of the steps should be determined by their functions and possible internal logic.

[0209] The method of the embodiment of the present application is described in detail above, and the device of the embodiment of the present application is provided below.

[0210] See also Figure 5 , Figure 5 This is a schematic diagram of the structure of a label prediction device provided in an embodiment of the present application. The label prediction device 1 includes: an acquisition unit 11, a prediction unit 12, and a mapping unit 13. Among them:

[0211] An acquisition unit 11 is used to acquire target features and a target model of a target object, wherein the target model is a model for predicting an interest tag of the object, wherein the interest tag is a tag of a material of interest to the object, and wherein the target model is obtained by training with training data, wherein the training data includes training features and training tags of the training object, and wherein the training tags represent the interest tags of the training object; wherein the training tags are predicted by using a language model, and the prediction reason for the training tags predicted by the language model satisfies a preset condition;

[0212] The prediction unit 12 is used to use the target feature as an input of the target model, so that the target model predicts potential interest tags of the target object based on the target feature, and the potential interest tags are tags of materials that the target object is interested in.

[0213] In combination with any embodiment of the present application, the label prediction device 1 further includes a mapping unit 13, which is used to:

[0214] Based on the similarity between the n preset tags and the potential interest tags, determining a target tag from the n preset tags;

[0215] The target tag is used to determine the recommended material for the target object.

[0216] In conjunction with any embodiment of the present application, the prediction unit 12 is further configured to:

[0217] Acquire an existing interested tag of the target object, where the existing interested tag is a tag of a material that the target object is interested in;

[0218] Using the target model to predict tags of materials that the target object is interested in based on the target features, and obtaining candidate tags of interest;

[0219] When the candidate interest tag is related to the existing interest tag and the candidate interest tag is different from the existing interest tag, the candidate interest tag is used as the potential interest tag.

[0220] In combination with any embodiment of the present application, the n preset tags include i parent tags and (ni) child tags of the i parent tags, and the parent tag corresponds to one or more of the child tags;

[0221] The mapping unit 13 is further used for:

[0222] Determine the tag with the highest similarity to the potential interest tag from the i parent tags to obtain a candidate tag;

[0223] When the similarity between the candidate tag and the potential interest tag is greater than or equal to a similarity threshold, taking the candidate tag as the target tag;

[0224] In the case where the similarity between the candidate tag and the potential interest tag is less than the similarity threshold, determining the subtag of the candidate tag from the (ni) subtags to obtain y candidate tags, where y is less than (ni);

[0225] Determine the tag with the highest similarity to the potential interest tag from the y candidate tags to obtain the target tag.

[0226] In combination with any embodiment of the present application, the mapping unit 13 is further configured to:

[0227] Acquire s reference tags of a reference object, wherein the reference tags are tags of materials of interest to the reference object, the reference tags are tags among the n preset tags, and s is a positive integer less than n;

[0228] Extracting an embedding of a reference feature of the reference object to obtain a reference feature embedding;

[0229] Extracting the embedding of the target feature of the target object to obtain a target feature embedding;

[0230] When the reference feature embedding matches the target feature embedding, a tag with the highest similarity to the potential interest tag is determined from the s reference tags to obtain the target tag.

[0231] In an embodiment of the present application, the training data includes training features and training labels of the training object, wherein the training label represents the label of the material of interest to the training object, and the training label is predicted by using a language model, and the prediction reason of the training label predicted by the language model satisfies a preset condition, so the accuracy of the training label is high. In this way, when the target model is trained by the training data, the target model can predict the interest label of the material of interest to the object based on the features of the object, and because the accuracy of the training label is high, the accuracy of the interest label predicted by the target model is high. Therefore, after obtaining the target features and the target model of the target object, the label prediction device uses the target features as the input of the target model, so that the target model can predict the potential interest labels of the material of interest to the target object based on the target features, and the accuracy of the potential interest labels can be improved.

[0232] See also Figure 6 , Figure 6 This is a schematic diagram of the structure of a label acquisition device provided in an embodiment of the present application. The label acquisition device 2 includes: an acquisition unit 21, a generation unit 22, a prediction unit 23, and a determination unit 24. Among them:

[0233] An acquisition unit 21 is used to acquire training features and a language model of a training object; the language model is a model for predicting interest tags of an object, and the interest tags are tags of materials that the object is interested in;

[0234] A generating unit 22, configured to generate a target prompt word according to the training feature, wherein the target prompt word is used to guide the language model to predict the label of the material of interest to the training subject according to the training feature, and to generate a reason for the predicted label;

[0235] A prediction unit 23, configured to predict the label of the material of interest to the training subject based on the training features using the language model under the guidance of the target prompt word, and obtain a predicted label;

[0236] The prediction unit 23 is further configured to generate a prediction reason for the prediction label using the language model under the guidance of the target prompt word;

[0237] The determination unit 24 is configured to use the predicted label as a training label when the prediction reason satisfies a preset condition, wherein the training label is a label of the material that the training subject is interested in.

[0238] In conjunction with any embodiment of the present application, the preset condition includes a confidence threshold;

[0239] The determination unit 24 is specifically configured to:

[0240] When the confidence of the prediction reason is greater than the confidence threshold, the prediction label is used as the training label.

[0241] In conjunction with any embodiment of the present application, the determining unit 24 is further configured to:

[0242] In a case where the prediction reason matches the training feature, it is determined that the confidence of the prediction reason is greater than the confidence threshold.

[0243] In combination with any embodiment of the present application, the target prompt word also includes a real interest tag of the training object;

[0244] The determination unit 24 is further configured to:

[0245] When the real interest tag is related to the predicted reason and the real interest tag is different from the predicted tag, it is determined that the confidence of the predicted reason is greater than the confidence threshold.

[0246] In the embodiment of the present application, the label acquisition device generates a target prompt word according to the training features after acquiring the training features and language model of the training object. Under the guidance of the target prompt word, the label acquisition device uses the language model to obtain the predicted label based on the training features and generates a predicted reason for the predicted label. When the predicted reason meets the preset conditions, the predicted label is determined to be the label of interest to the training object, that is, the training label. In this way, the labeling of the training features of the training object can be completed without manual labor, thereby reducing labor costs. Moreover, since the training label is obtained when the predicted reason meets the preset conditions, the accuracy of the training label is high.

[0247] See also Figure 7 , Figure 7This is a schematic diagram of the structure of another tag prediction device provided in an embodiment of the present application. The tag prediction device 3 includes: an acquisition unit 31, a generation unit 32, a prediction unit 33, and a determination unit 34. Among them:

[0248] An acquisition unit 31 is used to acquire target features and a prediction model of a target object, wherein the prediction model is a model for predicting an interest tag of the object, and the interest tag is a tag of a material that the object is interested in;

[0249] A generating unit 32, configured to generate a prediction prompt word according to the target feature, wherein the prediction prompt word is used to guide the prediction model to predict the label of the material of interest to the target object according to the target feature, and to generate a reason for the predicted label;

[0250] A prediction unit 33, configured to predict a label of the material that the target object is interested in based on the target feature by using the prediction model under the guidance of the prediction prompt word, and obtain a first label;

[0251] The prediction unit 33 is further configured to generate a first reason for predicting the first label using the prediction model under the guidance of the prediction prompt word;

[0252] The determination unit 34 is configured to determine a label of the material of interest to the target object based on the first label when it is determined based on the first reason that the confidence of the first label is greater than or equal to a label threshold.

[0253] In combination with any embodiment of the present application, the determination unit 34 is further configured to:

[0254] When the confidence of the first reason is greater than a confidence threshold, it is determined that the confidence of the first label is greater than or equal to the label threshold.

[0255] In conjunction with any embodiment of the present application, the determination unit 34 is specifically configured to:

[0256] In a case where the predicted reason matches the target feature, it is determined that the confidence of the first reason is greater than the confidence threshold.

[0257] In an embodiment of the present application, the label prediction device generates a prediction prompt word according to the target feature after obtaining the target feature and the prediction model of the target object, and uses the prediction prompt word as the input of the prediction model. Under the guidance of the prediction prompt word, the prediction model predicts the first label based on the target feature and generates a first reason for predicting the first label, and determines the confidence of the first label based on the first reason. When the confidence of the first label is greater than or equal to the threshold, the first label is determined to be the label of interest to the target object. In this way, the prediction of the label of the material of interest to the target object can be completed without manual labor, thereby reducing labor costs, and the confidence of the predicted first label is greater than or equal to the threshold, and the label is more accurate in being the label of the material of interest to the target object.

[0258] In some embodiments, the functions or modules included in the device provided in the embodiments of the present application can be used to execute the method described in the above method embodiments. The specific implementation can refer to the description of the above method embodiments. For the sake of brevity, it will not be repeated here.

[0259] Figure 8 A schematic diagram of the hardware structure of an electronic device provided in an embodiment of the present application. The electronic device 4 includes a processor 41 and a memory 42. Optionally, the electronic device 4 also includes an input device 43 and an output device 44. The processor 41, the memory 42, the input device 43 and the output device 44 are coupled via a connector, and the connector includes various interfaces, transmission lines or buses, etc., which are not limited in the embodiments of the present application. It should be understood that in each embodiment of the present application, coupling refers to mutual connection in a specific manner, including direct connection or indirect connection through other devices, for example, it can be connected through various interfaces, transmission lines, buses, etc.

[0260] The processor 41 may include one or more processors, for example, one or more central processing units (CPUs). When the processor is a CPU, the CPU may be a single-core CPU or a multi-core CPU. Optionally, the processor 41 may be a processor group consisting of multiple CPUs, and the multiple processors are coupled to each other through one or more buses. Optionally, the processor may also be other types of processors, etc., which are not limited in the embodiments of the present application.

[0261] The memory 42 can be used to store computer program instructions and various computer program codes including the program code for executing the program code of the present application. Optionally, the memory includes but is not limited to random access memory (RAM), read-only memory (ROM), erasable programmable read only memory (EPROM), or portable read only memory (CD-ROM), which is used for related instructions and data.

[0262] The input device 43 is used to input data and / or signals, and the output device 44 is used to output data and / or signals. The input device 43 and the output device 44 can be independent devices or an integrated device.

[0263] It can be understood that in the embodiment of the present application, the memory 42 can be used not only to store relevant instructions, but also to store relevant data. For example, the memory 42 can be used to store target features and target models obtained through the input device 43, or the memory 42 can also be used to store potential interest tags obtained through the processor 41, etc. The embodiment of the present application does not limit the specific data stored in the memory.

[0264] Understandably, Figure 8 Only a simplified design of an electronic device is shown. In practical applications, the electronic device may also include other necessary components, including but not limited to any number of input / output devices, processors, memories, etc., and all electronic devices that can implement the embodiments of the present application are within the protection scope of the present application.

[0265] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0266] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments, and will not be repeated here. Those skilled in the art can also clearly understand that the descriptions of the various embodiments of the present application have different focuses. For the convenience and brevity of description, the same or similar parts may not be repeated in different embodiments. Therefore, for parts not described or not described in detail in a certain embodiment, refer to the records of other embodiments.

[0267] In the several embodiments provided in the present application, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0268] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0269] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0270] In the above embodiments, it can be implemented in whole or in part by software, hardware, firmware or any combination thereof. When implemented using software, it can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the process or function described in the embodiment of the present application is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted through the computer-readable storage medium. The computer instructions can be transmitted from a website site, computer, server or data center to another website site, computer, server or data center by wired (e.g., coaxial cable, optical fiber, digital subscriber line (digital subscriber line, DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) mode. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more available media integrated. The available medium may be a magnetic medium (eg, a floppy disk, a hard disk, a magnetic tape), an optical medium (eg, a digital versatile disc (DVD)), or a semiconductor medium (eg, a solid state disk (SSD)).

[0271] A person skilled in the art can understand that all or part of the processes in the above-mentioned embodiments can be implemented by a computer program to instruct the relevant hardware, and the program can be stored in a computer-readable storage medium. When the program is executed, it can include the processes of the above-mentioned method embodiments. The aforementioned storage medium includes: a read-only memory (ROM) or a random access memory (RAM), a magnetic disk or an optical disk, and other media that can store program codes.

Claims

1. A label prediction method, characterized in that: The method comprises: Obtaining target features and a target model of a target object, wherein the target model is a model for predicting an interest tag of the object, wherein the interest tag is a tag of a material of interest to the object, and wherein the target model is obtained by training with training data, wherein the training data includes training features and training tags of the training object, and wherein the training tags represent tags of materials of interest to the training object; wherein the training tags are predicted by using a language model, and the prediction reason for the training tags predicted by the language model satisfies a preset condition; The target feature is used as an input of the target model, so that the target model predicts potential interest tags of the target object based on the target feature, and the potential interest tags are tags of materials that the target object is interested in.

2. The method according to claim 1, characterized in that After taking the target feature as an input of the target model so that the target model predicts the potential interest tag of the target object based on the target feature, the method further includes: Based on the similarity between the n preset tags and the potential interest tags, determining a target tag from the n preset tags; The target tag is used to determine the recommended material for the target object.

3. The method according to claim 1 or 2, characterized in that: The step of using the target feature as an input of the target model so that the target model predicts a potential interest tag of the target object based on the target feature, wherein the potential interest tag is a tag of a material that the target object is interested in, includes: Acquire an existing interested tag of the target object, where the existing interested tag is a tag of a material that the target object is interested in; Using the target model to predict tags of materials that the target object is interested in based on the target features, and obtaining candidate tags of interest; When the candidate interest tag is related to the existing interest tag and the candidate interest tag is different from the existing interest tag, the candidate interest tag is used as the potential interest tag.

4. The method according to claim 2, characterized in that: The n preset tags include i parent tags and (ni) child tags of the i parent tags, and the parent tag corresponds to one or more of the child tags; The determining a target tag from the n preset tags based on the similarity between the n preset tags and the potential interest tag comprises: Determine the tag with the highest similarity to the potential interest tag from the i parent tags to obtain a candidate tag; When the similarity between the candidate tag and the potential interest tag is greater than or equal to a similarity threshold, taking the candidate tag as the target tag; In the case where the similarity between the candidate tag and the potential interest tag is less than the similarity threshold, determining the subtag of the candidate tag from the (ni) subtags to obtain y candidate tags, where y is less than (ni); Determine the tag with the highest similarity to the potential interest tag from the y candidate tags to obtain the target tag.

5. The method according to claim 2, characterized in that: The determining a target tag from the n preset tags based on the similarity between the n preset tags and the potential interest tag comprises: Acquire s reference tags of a reference object, wherein the reference tags are tags of materials of interest to the reference object, the reference tags are tags among the n preset tags, and s is a positive integer less than n; Extracting an embedding of a reference feature of the reference object to obtain a reference feature embedding; Extracting the embedding of the target feature of the target object to obtain a target feature embedding; When the reference feature embedding matches the target feature embedding, a tag with the highest similarity to the potential interest tag is determined from the s reference tags to obtain the target tag.

6. A label acquisition method, characterized in that: The method comprises: Acquire training features and a language model of a training object; the language model is a model for predicting interest tags of the object, and the interest tags are tags of materials that the object is interested in; Generate a target prompt word according to the training features, wherein the target prompt word is used to guide the language model to predict the label of the material of interest to the training subject according to the training features, and to generate a reason for the predicted label; Under the guidance of the target prompt word, using the language model to predict the label of the material that the training subject is interested in based on the training features to obtain a predicted label; Under the guidance of the target prompt word, using the language model to generate a prediction reason for obtaining the predicted label; In the case where the prediction reason meets a preset condition, the prediction label is used as a training label, and the training label is a label of the material that the training object is interested in.

7. The method according to claim 6, characterized in that The preset conditions include a confidence threshold; When the prediction reason satisfies a preset condition, determining the prediction label as a training label includes: When the confidence of the prediction reason is greater than the confidence threshold, the prediction label is used as the training label.

8. The method according to claim 7, characterized in that In the case where the confidence of the prediction reason is greater than the confidence threshold, before using the prediction label as the training label, the method further includes: In a case where the prediction reason matches the training feature, it is determined that the confidence of the prediction reason is greater than the confidence threshold.

9. The method according to claim 7, characterized in that: The target prompt word also includes the real interest tag of the training object; When the confidence of the prediction reason is greater than the confidence threshold, before using the prediction label as the training label, the method further includes: When the real interest tag is related to the predicted reason and the real interest tag is different from the predicted tag, it is determined that the confidence of the predicted reason is greater than the confidence threshold.

10. A label prediction method, characterized in that: The method comprises: Obtaining target features and a prediction model of a target object, wherein the prediction model is a model for predicting an interest tag of the object, and the interest tag is a tag of a material of interest to the object; Generate a prediction prompt word according to the target feature, wherein the prediction prompt word is used to guide the prediction model to predict the label of the material of interest to the target object according to the target feature, and generate a reason for the predicted label; Under the guidance of the prediction prompt word, the prediction model is used to predict the label of the material that the target object is interested in based on the target feature to obtain a first label; Under the guidance of the prediction prompt word, using the prediction model to generate a first reason for predicting the first label; In a case where it is determined based on the first reason that the confidence of the first tag is greater than or equal to a tag threshold, a tag of the material of interest to the target object is determined based on the first tag.

11. The method according to claim 10, characterized in that In the case where it is determined based on the first reason that the confidence of the first tag is greater than or equal to a tag threshold, before determining the tag of the material of interest to the target object based on the first tag, the method further includes: When the confidence of the first reason is greater than a confidence threshold, it is determined that the confidence of the first label is greater than or equal to the label threshold.

12. The method according to claim 11, characterized in that In a case where the confidence of the first reason is greater than a confidence threshold, before determining that the confidence of the first tag is greater than or equal to the tag threshold, the method further includes: In a case where the predicted reason matches the target feature, it is determined that the confidence of the first reason is greater than the confidence threshold.

13. A label prediction device, characterized in that: The label prediction device comprises: an acquisition unit, configured to acquire a target feature and a target model of a target object, wherein the target model is a model for predicting an interest tag of the object, wherein the interest tag is a tag of a material of interest to the object, and wherein the target model is obtained by training with training data, wherein the training data includes training features and training tags of the training object, and wherein the training tags represent tags of materials of interest to the training object; wherein the training tags are obtained by prediction using a language model, and a prediction reason for the training tags predicted by the language model satisfies a preset condition; A prediction unit is used to use the target feature as an input of the target model so that the target model predicts a potential interest tag of the target object based on the target feature, wherein the potential interest tag is a tag of a material that the target object is interested in.

14. A label acquisition device, characterized in that: The label acquisition device comprises: An acquisition unit, used to acquire training features and a language model of a training object; the language model is a model for predicting interest tags of an object, and the interest tags are tags of materials that the object is interested in; A generating unit, configured to generate a target prompt word according to the training feature, wherein the target prompt word is used to guide the language model to predict a label of the material of interest to the training subject according to the training feature, and to generate a reason for the predicted label; A prediction unit, used to predict the label of the material of interest to the training subject based on the training features using the language model under the guidance of the target prompt word, to obtain a predicted label; The prediction unit is further configured to generate a prediction reason for the prediction label using the language model under the guidance of the target prompt word; A determination unit is used to use the predicted label as a training label when the prediction reason meets a preset condition, and the training label is a label of the material that the training object is interested in.

15. A label prediction device, characterized in that: The label prediction device comprises: An acquisition unit, used to acquire target features and a prediction model of a target object, wherein the prediction model is a model for predicting an interest tag of the object, and the interest tag is a tag of a material of interest to the object; A generating unit, configured to generate a prediction prompt word according to the target feature, wherein the prediction prompt word is used to guide the prediction model to predict the label of the material of interest to the target object according to the target feature, and to generate a reason for the predicted label; A prediction unit, configured to predict a label of the material of interest to the target object based on the target feature using the prediction model under the guidance of the prediction prompt word, to obtain a first label; The prediction unit is further configured to generate a first reason for predicting the first label using the prediction model under the guidance of the prediction prompt word; A determination unit is configured to determine a label of the material of interest to the target object based on the first label when it is determined based on the first reason that the confidence of the first label is greater than or equal to a label threshold.

16. An electronic device, characterized in that: include: A processor and a memory, wherein the memory is used to store computer program codes, wherein the computer program codes include computer instructions, and when the processor executes the computer instructions, the electronic device executes the method as claimed in any one of claims 1 to 5, or executes the method as claimed in any one of claims 6 to 9, or executes the method as claimed in any one of claims 10 to 12.

17. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, which includes program instructions. When the program instructions are executed by a processor, the processor is caused to execute the method described in any one of claims 1 to 5, or the method described in any one of claims 6 to 9, or the method described in any one of claims 10 to 12.

18. A computer program product, characterized in that The computer program product includes a computer program or instructions; when the computer program or instructions are run on a computer, the computer is enabled to execute the method described in any one of claims 1 to 5, or the method described in any one of claims 6 to 9, or the method described in any one of claims 10 to 12.

Images