Information promotion method and related equipment

Abstract

The invention provides an information promotion method. The method comprises the steps of determining predicted sales volume of each cold start object in a cold start object pool based on a sales volume prediction model; wherein the sales prediction model is a large language model subjected to supervised fine tuning; determining at least one candidate cold start object based on the predicted sales volume of each cold start object; selecting a target object for information promotion from the at least one candidate cold start object and the candidate non-cold start object; and popularizing the target object. The invention further provides a corresponding information popularization related device.

Description

Technical Field

[0001] The present disclosure relates to the field of computer technologies, and particularly to an information promotion method and related devices. Background Art

[0002] Currently, e-commerce platforms usually face difficulties in screening a large number of global network products when introducing new products. For example, global network product information usually contains a large amount of unstructured text data, such as product titles, descriptions, etc., and lacks statistical data such as sales volume. As a result, the accuracy and effectiveness of screening global network products are both relatively low. However, predicting product sales volume also has considerable difficulties, especially in the scenario of products with long-tail distribution, where the deviation of product sales volume prediction is usually large. Summary of the Invention

[0003] In view of this, embodiments of the present disclosure provide an information promotion method, which can improve the sales volume prediction ability for cold start objects by performing supervised fine-tuning on a large language model, help cold start objects quickly break through the cold start period, improve the performance of cold start objects in e-commerce platforms, and thus enhance the user experience.

[0004] The information promotion method described in embodiments of the present disclosure may include: respectively determining the predicted sales volume of each cold start object in a cold start object pool based on a sales volume prediction model; wherein, the sales volume prediction model is a large language model that has undergone supervised fine-tuning; determining at least one candidate cold start object based on the predicted sales volume of each cold start object; selecting a target object for information promotion from the at least one candidate cold start object and candidate non-cold start objects; and promoting the target object.

[0005] The information promotion method described in embodiments of the present disclosure may further include: constructing a training data set; wherein, the training data set includes: a plurality of training samples and discrete sales volume labels corresponding to the training samples; generating sales volume prediction prompt words for each training sample in the training data set; inputting the sales volume prediction prompt words into the large language model; obtaining a plurality of predicted sales volumes respectively corresponding to the plurality of training samples output by the large language model; respectively determining the loss between the predicted sales volume corresponding to each training sample and the discrete sales volume label corresponding to the training sample; and performing supervised fine-tuning on the large language model based on the loss to obtain the large language model that has undergone supervised fine-tuning.

[0006] The construction of the training dataset in the embodiments of the present disclosure includes: respectively annotating the corresponding sales volume values for multiple samples in a preset sample dataset; for samples with a sales volume annotation value of zero, setting the discrete sales volume label corresponding to the sample to zero; for samples with a non-zero sales volume annotation value, dividing the samples into N sample groups according to the preset number of groups N and the N quantiles where the sales volume annotation value of the sample is located, and respectively setting the same discrete sales volume label for the samples within each sample group; wherein, the larger the sales volume annotation value of the sample, the larger the discrete sales volume label set for the sample group where the sample is located; N is a positive integer; and respectively for each discrete sales volume label, extracting a predetermined number of samples from the samples corresponding to the discrete sales volume label as the training samples.

[0007] In the embodiments of the present disclosure, generating a sales volume prediction prompt word for each training sample in the training dataset includes: generating the sales volume prediction prompt word based on a preset sales volume prediction prompt word template and the feature information of the training sample; wherein, the feature information of the training sample includes: one or any combination of the basic information of the object, the picture information of the object, the stock keeping unit (SKU) configuration information of the object, the price information of the object, and the information of the seller of the object; the preset sales volume prediction prompt word template includes: a role description part, a background information part, an input data description part, an output requirement part, and a special instruction part.

[0008] In the embodiments of the present disclosure, determining at least one candidate cold start object based on the predicted sales volumes of the respective cold start objects includes: adding at least one cold start object with a predicted sales volume greater than a preset sales volume threshold to a cold start object recall pool; performing a rough ranking on the cold start objects in the cold start object recall pool based on the feature information of the cold start objects, and selecting a predetermined number of cold start objects from the results of the rough ranking and adding them to a fine ranking pool; performing a fine ranking on the cold start objects in the fine ranking pool based on the feature information of the cold start objects; and determining the at least one candidate cold start object from the cold start objects in the fine ranking pool according to the results of the fine ranking.

[0009] In the embodiments of the present disclosure, determining at least one candidate cold start object based on the predicted sales volumes of the respective cold start objects includes: adding at least one cold start object with a predicted sales volume greater than a preset sales volume threshold to a cold start object recall pool; performing a rough ranking on the cold start objects in the cold start object recall pool based on the feature information of the cold start objects and the predicted sales volumes of the cold start objects, and selecting a predetermined number of cold start objects from the results of the rough ranking and adding them to a fine ranking pool; performing a fine ranking on the cold start objects in the fine ranking pool based on the feature information of the cold start objects; and determining the at least one candidate cold start object from the cold start objects in the fine ranking pool according to the results of the fine ranking.

[0010] In the embodiments of the present disclosure, determining at least one candidate cold start object based on the predicted sales volumes of the respective cold start objects includes: adding at least one cold start object with a predicted sales volume greater than a preset sales volume threshold to a cold start object recall pool; performing a rough ranking on the cold start objects in the cold start object recall pool based on the feature information of the cold start objects and the predicted sales volumes of the cold start objects, and selecting a predetermined number of cold start objects from the rough ranking results to be added to a fine ranking pool; performing a fine ranking on the cold start objects in the fine ranking pool based on the feature information of the cold start objects and the predicted sales volumes of the cold start objects; and determining the at least one candidate cold start object from the cold start objects in the fine ranking pool according to the fine ranking results.

[0011] In the embodiments of the present disclosure, performing a rough ranking on the cold start objects in the cold start object recall pool based on the feature information of the cold start objects and the predicted sales volumes of the cold start objects includes: performing the following operations respectively for each cold start object in the cold start object recall pool: determining a rough ranking feature index of the cold start object based on the feature information of the cold start object; performing a weighted sum of the predicted sales volume of the cold start object and the feature index of the cold start object to obtain a rough ranking comprehensive index of the cold start object; wherein, the sum of the weight coefficient of the predicted sales volume of the cold start object and the weight coefficient of the rough ranking feature index of the cold start object is equal to 1; and performing a rough ranking on the cold start objects in the cold start object recall pool based on the rough ranking comprehensive index of the cold start object.

[0012] In the embodiments of the present disclosure, performing a fine ranking on the cold start objects in the fine ranking pool based on the feature information of the cold start objects and the predicted sales volumes of the cold start objects includes: determining a plurality of fine ranking features based on the feature information of the cold start objects and the predicted sales volumes of the cold start objects; wherein, taking the predicted sales volume of the cold start object as one of the plurality of fine ranking features; and performing a fine ranking on the cold start objects in the fine ranking pool based on the plurality of fine ranking features.

[0013] In the embodiments of the present disclosure, determining a target object for information promotion from the at least one candidate cold start object and the determined candidate non-cold start objects includes: performing a mixed ranking on the at least one candidate cold start object and the candidate non-cold start objects; and determining the target object based on the mixed ranking results.

[0014] Corresponding to the above information promotion method, an embodiment of the present disclosure also discloses an information promotion device, including:

[0015] A sales volume prediction module, configured to respectively determine the predicted sales volumes of the respective cold start objects in a cold start object pool based on a sales volume prediction model; wherein, the sales volume prediction model is a large language model that has been supervised and fine-tuned.

[0016] A candidate cold start object determination module, configured to determine at least one candidate cold start object based on the predicted sales volumes of the respective cold start objects;

[0017] A target object determination module, configured to select a target object for information promotion from the at least one candidate cold start object and candidate non-cold start objects; and

[0018] An information promotion module, configured to promote the target object.

[0019] In addition, an embodiment of the present disclosure further provides an electronic device, including: a memory, a processor, and a computer program stored on the memory and executable on the processor, where when the processor executes the program, the above information promotion method is implemented.

[0020] An embodiment of the present disclosure further provides a non-transitory computer-readable storage medium, where the non-transitory computer-readable storage medium stores computer instructions, and the computer instructions are used to cause a computer to execute the above information promotion method.

[0021] An embodiment of the present disclosure further provides a computer program product, including computer program instructions, where when the computer program instructions run on a computer, the computer is caused to execute the above information promotion method.

[0022] It can be seen from this that the above information promotion method and related devices can obtain a sales volume prediction model capable of accurately predicting the sales volumes of cold start objects through supervised fine-tuning of a large language model, and select at least one candidate cold start object from the cold start object pool based on the sales volumes predicted by the above sales volume prediction model, and participate with other candidate non-cold start objects in determining the target object for information promotion, so as to promote the target object. It can be seen that the above solution helps cold start objects quickly break through the cold start period and improve the performance of cold start objects on e-commerce platforms by improving the sales volume prediction ability for cold start objects, and also greatly improves the user experience. Description of the Drawings

[0023] In order to more clearly illustrate the technical solutions in the present disclosure or related technologies, the following will briefly introduce the drawings required for use in the embodiments or related technology descriptions. Obviously, the drawings in the following descriptions are only embodiments of the present disclosure. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0024] Figure 1 Shows the implementation process of the information promotion method described in some embodiments of the present disclosure.

[0025] Figure 2Shows the implementation process of the method for supervised fine-tuning of large language models according to some embodiments of the present disclosure.

[0026] Figure 3 Shows the implementation process of the specific method for constructing a training dataset according to some embodiments of the present disclosure.

[0027] Figure 4 Shows the method for determining at least one candidate cold start object from a cold start object pool based on the predicted sales volume of a cold start object according to some embodiments of the present disclosure.

[0028] Figure 5 Shows the method for determining at least one candidate cold start object from a cold start object pool based on the predicted sales volume of a cold start object according to some other embodiments of the present disclosure.

[0029] Figure 6 Shows the method for determining at least one candidate cold start object from a cold start object pool based on the predicted sales volume of a cold start object according to some further embodiments of the present disclosure.

[0030] Figure 7 Shows the internal structure of the information promotion device according to some embodiments of the present disclosure.

[0031] Figure 8 Illustrates a more specific schematic diagram of the hardware structure of an electronic device according to some embodiments of the present disclosure. Detailed implementation manners

[0032] To make the objectives, technical solutions, and advantages of the present disclosure clearer and more understandable, the following further describes the present disclosure in detail with reference to specific embodiments and the accompanying drawings.

[0033] It should be noted that unless otherwise defined, the technical terms or scientific terms used in the embodiments of the present disclosure should have the ordinary meanings understood by those of ordinary skill in the art to which the present disclosure belongs. The "first", "second", and similar terms used in the embodiments of the present disclosure do not indicate any order, quantity, or importance, but are only used to distinguish different components. The terms such as "including" or "comprising" mean that the elements or objects appearing before this word cover the elements or objects listed after this word and their equivalents, without excluding other elements or objects. The terms such as "connected" or "coupled" are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. The terms such as "upper", "lower", "left", "right", etc. are only used to represent relative positional relationships, and when the absolute position of the object being described changes, the relative positional relationship may also change accordingly.

[0034] It is understandable that before using the technical solutions of the various embodiments of the present disclosure, the types, usage scopes, usage scenarios, etc. of the personal information involved will be informed to the user in an appropriate manner, and the user's authorization will be obtained.

[0035] For example, when responding to receiving an active request from a user, a prompt message is sent to the user to clearly prompt the user that the operation requested by the user will require obtaining and using the user's personal information. Thus, the user can autonomously choose whether to provide personal information to software or hardware such as an electronic device, an application program, a server, or a storage medium that performs the operations of the technical solutions of the present disclosure according to the prompt message.

[0036] As an optional but non-limiting implementation manner, the manner of sending a prompt message to the user in response to receiving an active request from the user may be, for example, in the form of a pop-up window, and the prompt message may be presented in text in the pop-up window. In addition, the pop-up window may also carry a selection control for the user to choose "agree" or "disagree" to provide personal information to the electronic device.

[0037] It is understandable that the above process of notifying and obtaining the user's authorization is only illustrative and does not constitute a limitation on the implementation manner of the present disclosure. Other manners that meet relevant laws and regulations can also be applied to the implementation manner of the present disclosure.

[0038] For the sake of clarity in description, before describing the specific technical solutions of the embodiments of the present disclosure, several technical terms involved in the embodiments of the present disclosure are first described.

[0039] A large language model (LLM, Large Language Model) is a neural network model used for natural language processing tasks and having a large number of parameters and a complex structure. The LLM is trained by using a large amount of data and uses deep learning techniques to learn the basic patterns and structures of language. Compared with small models, the LLM can achieve higher performance and usually has better generalization in various tasks.

[0040] Supervised Fine Tuning (SFT) refers to the process of fine-tuning a pre-trained neural network model using new data with labels to improve its performance.

[0041] A prompt is an injection instruction used to "command" a large model to think about problems and output content according to a preset idea. A prompt is an instruction or information that guides or triggers a large model to make a response.

[0042] An object, in the embodiments of the present disclosure, refers to a commodity, a product, or an information content involved in an e-commerce platform.

[0043] A cold start object refers to a commodity or product newly added to an e-commerce platform. Usually, the sales volume corresponding to a cold start object is usually zero.

[0044] A cold start object pool refers to a set composed of all cold start objects on an e-commerce platform.

[0045] A non-cold start object refers to a commodity, product, or information content involved in an e-commerce platform; among them, for non-cold start commodities or products, their sales volume is usually not zero.

[0046] Information promotion refers to exposing some or all of the characteristic information of the above objects for users of the e-commerce platform to browse.

[0047] As mentioned above, when introducing new products, an e-commerce platform usually faces the problem of difficulty in screening a large number of commodities across the network. For this reason, the embodiments of the present disclosure provide an information promotion method.

[0048] Figure 1 Shows the implementation process of the information promotion method described in the embodiments of the present disclosure. As Figure 1 shown, the information promotion method described in the embodiments of the present disclosure may include the following multiple steps:

[0049] In step 110, based on a sales volume prediction model, determine the predicted sales volume of each cold start object in the cold start object pool. In the embodiments of the present disclosure, the above sales volume prediction model is a large language model that has been fine-tuned with supervision.

[0050] In step 120, determine at least one candidate cold start object based on the predicted sales volume of each of the above cold start objects.

[0051] In step 130, select target objects for information promotion from the above at least one candidate cold start object and candidate non-cold start objects.

[0052] In step 140, promote the above target objects.

[0053] It can be seen from this that in the embodiments of the present disclosure, during the process of information promotion, the sales volume of each cold start object in the cold start object pool can be predicted first using a sales volume prediction model, and based on the predicted sales volume output by the sales volume prediction model, cold start objects with sales potential are selected from the cold start objects and added to the target objects for information promotion, thereby helping the cold start objects quickly break through the cold start period, improving the performance of the cold start objects in the e-commerce platform, and also greatly improving the user experience.

[0054] The following will specifically illustrate the specific implementation methods of each step in the embodiments of the present disclosure with specific examples.

[0055] As described above, in the above step 110, the above sales volume prediction model is a large language model that has been supervised and fine-tuned. Next, the specific method for supervised fine-tuning of the large language model in the embodiments of the present disclosure will be described in detail with reference to specific examples.

[0056] Figure 2 shows the implementation process of the method for supervised fine-tuning of the large language model described in the embodiments of the present disclosure. As Figure 2 shown, the specific method for supervised fine-tuning of the large language model includes:

[0057] In step 210, a training data set is constructed.

[0058] In the embodiments of the present disclosure, the above training data set may include a plurality of training samples and discrete sales volume labels corresponding to the training samples. Specifically, considering the long-tail distribution and class imbalance problems of sales volume values of various commodities on the e-commerce platform, the embodiments of the present disclosure first transform the sales volume prediction task into a sales volume scoring task, that is, discrete sales volume labels are used instead of actual sales volume values to represent the sales volume of an object, thereby solving the challenges of the traditional sales volume prediction model in the face of long-tail distribution and difficulty in sales volume order consistency. Specifically, in the embodiments of the present disclosure, first, the sales volume values corresponding to each training sample are discretely quantized into a specific value within 0 to A (for example, a scoring interval of 0 to 9 or 0 to 10), where the higher the value, the higher the sales volume corresponding to the training sample, thereby avoiding the instability and model bias that may be brought about when directly predicting sales volume values. That is to say, in the embodiments of the present disclosure, the above discrete sales volume label is the value obtained by quantizing the sales volume value of the training sample.

[0059] Based on the above content, in the embodiments of the present disclosure, the specific method for constructing the training data set can be as Figure 3 shown, including the following multiple steps:

[0060] In step 310, the corresponding sales volume values of multiple samples in the pre-set sample data set are respectively labeled.

[0061] In the embodiments of the present disclosure, the above sample may correspond to one of the aforementioned objects and include various characteristic information of the above object. In the embodiments of the present disclosure, the above characteristic information may include: object basic information, picture information of the object, minimum stock unit (SKU) configuration information of the object, price information of the object, and seller information of the object, etc., one or any combination thereof. The above characteristic information will be used for supervised fine-tuning of the large language model, and the goal is to enable the large language model to have a deeper understanding of the characteristics of the sample.

[0062] Specifically, the basic information of the above object may include one or any combination of the following information:

[0063] Title of the object: Reflects the main information of the object and can be used as the core text input for the large language model;

[0064] Category of the object: Helps the large language model understand the classification and uses of the object;

[0065] Description information of the object: The large language model can extract and analyze key phrases and important information in the object description information to support feature learning.

[0066] The picture information of the above object may include one or any combination of the following information:

[0067] Address information (URL) of the main picture: One of the important features of the object;

[0068] Picture list: Includes the URLs and their order of all associated pictures of the corresponding object, used to enhance the large language model's comprehensive understanding of the object display;

[0069] Picture position: Used to guide the large language model to judge the visual display logic of the object in combination with the picture order.

[0070] The SKU configuration information of the above object may include one or any combination of the following information:

[0071] Number of SKUs: Reflects the inventory and variety diversity of the object;

[0072] SKU size information: Used to guide the large language model to analyze the specific configurations available for the object (such as size and color, etc.).

[0073] The price information of the above object may include one or any combination of the following information:

[0074] Lowest SKU price: Provides the lowest price of the object, serving as the lower bound of pricing;

[0075] Highest SKU price: Provides the highest price of the object, reflecting the price range;

[0076] Price change: Used to guide the large language model to capture the impact of price elasticity and promotional activities through historical price fluctuation trends.

[0077] The seller information of the above object may include one or any combination of the following information:

[0078] Main business category: Used to guide the large language model to locate the main categories and their specialization levels in which the seller operates.

[0079] Description: Used to guide the large language model to analyze the seller's description to identify its service characteristics.

[0080] Multi-platform entry situation: Used to guide the large language model to evaluate the market coverage breadth of the seller and its operation performance on different platforms.

[0081] In addition, in the embodiments of the present disclosure, the annotation described in step 310 above can be implemented by manual annotation or other annotation methods. The above annotation process can be obtained through the analysis of the historical sales data and trends of the object samples, so as to ensure the high quality and consistency of the data.

[0082] In addition, it should be noted that for the above-mentioned pre-set sample data set, before performing the data annotation described in step 310 above, data cleaning needs to be carried out first, removing the outliers therein, and performing deduplication processing on the data, etc., so as to ensure the cleanliness and efficiency of the data in the sample data set.

[0083] In step 320, for the samples with a sales annotation value of zero, set the discrete sales label corresponding to the above samples to zero.

[0084] As mentioned above, in the embodiments of the present disclosure, the samples with a sales annotation value of zero can be considered as the samples of the cold start objects described in the embodiments of the present disclosure.

[0085] In step 330, for the samples with a sales annotation value not equal to zero, divide the samples into N sample groups according to the pre-set number of groups N and the N quantiles to which the sales annotation value of the sample belongs, and set the same discrete sales label for each sample within each sample group; among them, the larger the sales annotation value of the sample, the larger the discrete sales label set for the sample group where the sample is located; and N is a positive integer.

[0086] For example, in a specific example, it can be set that the above discrete sales labels include a total of eleven values from 0 to 10. Among them, when the discrete sales label is 0, it represents that the sales value of the training sample (cold start object sample) is 0. And the sales values of the training samples corresponding to the discrete sales labels 1 to 10 are all not 0 (non-cold start object samples). Specifically, when the above discrete sales label is 1 to 10, it represents the number of the quantiles in the deciles where the sales volume of the training sample is located. For example, the discrete sales label of 10 represents that the sales volume of the training sample belongs to the highest top 10%; the discrete sales label of 9 represents that the sales volume of the training sample belongs to the higher 11% - 20%;...; the discrete sales label of 1 represents that the sales volume of the training sample belongs to the lowest 10%. That is to say, the larger the value corresponding to the discrete sales label, the larger the sales value. The specific values of the above discrete sales labels are only examples. In the embodiments of the present disclosure, other N values can be selected and the above discrete sales labels can be determined based on the selected N value.

[0087] It can be understood that the above method of setting discrete sales volume labels ensures the stability of the sales volume prediction model when dealing with the problems of sales volume value fluctuations and unevenness, and avoids model biases that may be caused by sales volume value fluctuations and the long-tail effect.

[0088] In step 340, for each discrete sales volume label, a predetermined number of samples are respectively drawn from the samples corresponding to the discrete sales volume label as training samples.

[0089] Considering that the sales volume of many samples is low, if the large language model is supervised and fine-tuned based on all samples, it will be difficult for the large language model to fully learn the relevant features of high-sales volume objects. To solve this problem, in the embodiments of the present disclosure, for each discrete sales volume label, a predetermined number of samples are respectively drawn from the samples corresponding to the discrete sales volume label as training samples. Among them, for samples with non-zero discrete sales volume labels (i.e., moving sales objects), stratified uniform sampling can be performed. For example, a substantially same predetermined number of samples are set for sampling for samples with non-zero discrete sales volume labels, that is, for each discrete sales volume label, the same number of samples are respectively drawn. For example, in the above example, 5000 samples can be respectively selected from the training samples with discrete sales volume labels of 1 to 10 as the above training samples. By this way of stratified uniform sampling, the information of high-sales volume samples can be enhanced, so that the large language model can learn more relevant features of high-sales volume samples and improve its prediction ability. And to improve the discrimination ability of the large language model, especially the prediction ability for cold start objects (non-moving sales objects, that is, discrete sales volume labels are zero), samples of the same order of magnitude can also be added as negative samples for the supervised fine-tuning of the large language model. For example, in the above example, in addition to respectively selecting 5000 samples from the training samples with discrete sales volume labels of 1 to 10 as the above training samples, 5000 to 10000 samples are further selected from the training samples with discrete sales volume labels of 0 and added to the above training samples. This approach can effectively balance the positive and negative class ratios of the samples and enhance the recognition and prediction ability of the large language model for cold start samples.

[0090] It can be understood that by the above method of sampling samples according to the quantiles of sales volume values and discretizing sales volume labels, the distribution of training samples in the training data set can be greatly optimized. After the large language model is supervised and fine-tuned based on such a training data set, the rank correlation and prediction accuracy of the large language model can be effectively improved.

[0091] In step 220, a sales volume prediction prompt word is respectively generated for each training sample in the above training data set.

[0092] In an embodiment of the present disclosure, step 220 may specifically include: generating a sales volume prediction prompt word based on a pre-set sales volume prediction prompt word template and the feature information of the training samples; wherein, the pre-set sales volume prediction prompt word template may include: a role description part, a background information part, an input data description part, an output requirement part, and a special instruction part. Among them, the role description part is used to inform the large language model of its identity when performing the sales volume prediction task. For example, it informs the large language model that as an e-commerce platform operator, it is responsible for predicting the sales volume score of an object across the network based on the object's data. The background information part is mainly used to introduce the background and related information of the sales volume prediction task to the large language model. The input data description part is mainly used to provide the feature information of the training samples to the large language model, such as including one or any combination of the aforementioned object basic information, object picture information, object SKU configuration information, object price information, and object seller information, etc. In a specific application, these feature information can be provided in the JavaScript Object Notation (JSON) format. The output requirement part is mainly used to inform the large language model of the specific requirements for the output. The special instruction part is mainly used to prompt the large language model about the processing methods in case of special situations, such as the processing or representation method when there are missing fields in the data, the special relationship between data fields, and how to verify the output result, etc. It should be noted that the above sales volume prediction prompt word template is only an example, and in actual applications, the sales volume prediction prompt word template can be adjusted in any form according to actual needs.

[0093] In step 230, input the above sales volume prediction prompt word into the large language model.

[0094] In step 240, obtain multiple predicted sales volumes corresponding to the above multiple training samples respectively output by the above large language model.

[0095] Here, according to the indication of the specific requirements for the output in the above sales volume prediction prompt word template, the predicted sales volume output by the large language model will also be one of multiple pre-set discrete values, that is, the score for the sales volume of the training sample.

[0096] In step 250, for each training sample, respectively determine the loss between the predicted sales volume corresponding to the training sample and the discrete sales volume label corresponding to the training sample.

[0097] In step 260, perform supervised fine-tuning on the above large language model based on the above loss to obtain the above large language model after supervised fine-tuning.

[0098] In the embodiments of the present disclosure, various specific training methods can be adopted, for example, the Low-Rank Adaptation (LoRA) algorithm, etc., to achieve the above-mentioned supervised fine-tuning of the large language model.

[0099] It can be seen from this that the above method can perform supervised fine-tuning on the large language model based on the discrete sales volume labels corresponding to the training samples, so as to achieve the prediction and ranking of sales volume. In addition, in the process of the above-mentioned supervised fine-tuning of the large language model, multi-dimensional features such as text and images in the training samples can also be combined to further optimize the performance of the large language model after supervised fine-tuning for sales volume prediction.

[0100] After the large language model is supervised and fine-tuned by the above method, a sales volume prediction model can be obtained. Thus, after the e-commerce platform receives an information promotion request from a user, it can start to execute the above step 110, that is, based on the sales volume prediction model obtained through the above-mentioned supervised fine-tuning, determine the predicted sales volume of each cold start object in the cold start object pool. It can be understood that the above predicted sales volume is also one of multiple discrete values, that is, the score of the predicted sales volume. For example, the value of the predicted sales volume is one of multiple values from 0 to A.

[0101] Next, in the above step 120, at least one candidate cold start object can be determined from each cold start object in the cold start object pool based on the predicted sales volume of each cold start object output by the above sales volume prediction model.

[0102] It can be understood that generally, the implementation process of information promotion can include three processes: recall, rough ranking, and fine ranking. Among them, in the embodiments of the present disclosure, the predicted sales volume of the above cold start objects can be applied to one or more of the above recall, rough ranking, or fine ranking processes, so as to effectively expose the cold start objects in multiple ways.

[0103] In some embodiments of the present disclosure, the predicted sales volume of the above cold start objects can be only applied to the above recall process. In this case, the above step 120 can be implemented through, for example Figure 4 the process shown. As Figure 4 shown, the method for determining at least one candidate cold start object from each cold start object in the cold start object pool based on the predicted sales volume of each cold start object output by the sales volume prediction model can include the following multiple steps:

[0104] In step 410, add at least one cold start object whose predicted sales volume is greater than a pre-set sales volume threshold to the cold start object recall pool.

[0105] As described above, in the embodiments of the present disclosure, the larger the value of the predicted sales volume, the larger the corresponding predicted sales volume. For cold start objects with a predicted sales volume greater than a pre-set sales volume threshold, they are considered cold start objects with high sales potential and can be preferentially recommended. Therefore, these cold start objects with high sales potential can be added to an object pool for subsequent rough ranking and fine ranking. In the embodiments of the present disclosure, this object pool is called the cold start object recall pool. In addition, the above-mentioned pre-set sales volume threshold can be flexibly set according to the actual situation. It can be understood that the smaller the sales volume threshold, the relatively higher the probability of finally regarding the cold start object as the target object (the lower the threshold); and the larger the sales volume threshold, the relatively lower the probability of finally regarding the cold start object as the target object (the higher the threshold).

[0106] In step 420, based on the feature information of the cold start objects, rough ranking is performed on the cold start objects in the cold start object recall pool, and a predetermined number of cold start objects are selected from the rough ranking results and added to the fine ranking pool.

[0107] In step 430, fine ranking is performed on the cold start objects in the fine ranking pool based on the above-mentioned feature information of the cold start objects.

[0108] In step 440, at least one candidate cold start object is determined from the cold start objects in the fine ranking pool according to the fine ranking results.

[0109] It should be noted that the specific implementation methods of the above-mentioned rough ranking and fine ranking can refer to the rough ranking and fine ranking algorithms used by existing recommendation platforms for information promotion, and will not be repeated here.

[0110] It can be seen from this that in the above embodiments, cold start objects with a predicted sales volume greater than a pre-set sales volume threshold can be added to a separate path of object recall, adding an exclusive recall quota for cold start objects, thereby increasing the exposure rate of cold start objects with high sales potential, and thus improving the new object introduction mechanism.

[0111] In some other embodiments of the present disclosure, the predicted sales volume of the above-mentioned cold start objects can be applied to the above-mentioned recall process and rough ranking process. In this case, the above step 120 can be implemented through, for example Figure 5 the process shown. As Figure 5 shown, the method for determining at least one candidate cold start object from each cold start object in the cold start object pool based on the predicted sales volume of each cold start object output by the sales volume prediction model may include the following multiple steps:

[0112] In step 510, at least one cold start object with a predicted sales volume greater than a pre-set sales volume threshold is added to the cold start object recall pool.

[0113] The specific implementation method of the above step 510 is the same as that of the above step 410, and will not be repeated here.

[0114] In step 520, based on the feature information of the cold start object and the predicted sales volume of the above cold start object, the cold start objects in the cold start object recall pool are roughly ranked, and a predetermined number of cold start objects are selected from the rough ranking results and added to the fine ranking pool.

[0115] Specifically, in the embodiments of the present disclosure, in the above step 520, the following operations can be respectively performed for each cold start object in the cold start object recall pool: First, determine the rough ranking feature index of the cold start object based on the feature information of the cold start object; and perform a weighted sum of the predicted sales volume of the above cold start object and the feature index of the above cold start object to obtain the rough ranking comprehensive index of the above cold start object; wherein, the sum of the weight coefficients of the predicted sales volume of the cold start object and the weight coefficient of the rough ranking feature index of the cold start object is equal to 1; finally, based on the rough ranking comprehensive index of the cold start object, the cold start objects in the above cold start object recall pool are roughly ranked.

[0116] Specifically, the above determination of the rough ranking feature index of the cold start object based on the feature information of the cold start object can specifically be the rough ranking feature index of the cold start object determined based on the feature information of the cold start object by using an existing rough ranking algorithm, that is, the rough ranking score value Score1 obtained after performing a conventional rough ranking on the cold start object. In addition, the predicted sales volume of the cold start object is denoted as Score2 and used as the sales volume score of the cold start object. Next, perform a weighted sum of the above rough ranking feature index and the predicted sales volume: Score rough ranking = α·Score2+(1-α)·Score1, so as to obtain the rough ranking comprehensive index Score rough ranking of the above cold start object, that is, the comprehensive score of the rough ranking. Finally, based on the above rough ranking comprehensive index, the above cold start object is sorted, so as to obtain the rough ranking result. Wherein, α is the weight coefficient of the predicted sales volume and can be dynamically adjusted according to the cold start target (its initial value can be set to 0.2 - 0.3 according to the actual application situation).

[0117] In step 530, based on the feature information of the cold start object, the cold start objects in the fine ranking pool are finely ranked.

[0118] It should be noted that the specific implementation method of the above fine ranking can refer to the fine ranking algorithm used when an existing recommendation platform promotes information, and will not be repeated here.

[0119] In step 540, at least one candidate cold start object is determined from the cold start objects in the fine ranking pool according to the fine ranking result.

[0120] It can be seen from the above that in the above embodiments, cold start objects with predicted sales volume greater than a pre-set sales volume threshold can be used as one path for object recall, and the priority of objects with high predicted sales volume in the rough ranking of cold start objects can be improved through a weighting mechanism, so as to increase the exclusive recall quota for cold start objects with high sales potential, ensure that cold start objects have a certain competitiveness, and further increase the exposure rate of cold start objects with high sales potential, thereby improving the mechanism for introducing new objects. In addition, in the embodiments of the present disclosure, controlling α will not significantly interfere with the overall rough ranking distribution of mainstream objects.

[0121] In some other embodiments of the present disclosure, the predicted sales volume of the above cold start objects can be applied to the above recall process, rough ranking process, and fine ranking process. In this case, the above step 120 can be implemented by, for example Figure 6 the process shown. As Figure 6 shown, the method for determining at least one candidate cold start object from each cold start object in the cold start object pool based on the predicted sales volume of each cold start object output by the sales volume prediction model may include the following multiple steps:

[0122] In step 610, at least one cold start object with a predicted sales volume greater than a pre-set sales volume threshold is added to the cold start object recall pool.

[0123] The specific implementation method of the above step 610 is the same as the above step 410, and will not be repeated here.

[0124] In step 620, the cold start objects in the cold start object recall pool are roughly ranked based on the feature information of the cold start objects and the predicted sales volume of the cold start objects, and a predetermined number of cold start objects are selected from the rough ranking results and added to the fine ranking pool.

[0125] The specific implementation method of the above step 620 is the same as the above step 520, and will not be repeated here.

[0126] In step 630, the cold start objects in the fine ranking pool are finely ranked based on the feature information of the cold start objects and the predicted sales volume of the cold start objects.

[0127] In an embodiment of the present disclosure, step 630 may specifically include: determining a plurality of fine-rank features based on the feature information of the cold start object and the predicted sales volume of the cold start object; wherein, taking the predicted sales volume of the cold start object as one of the plurality of fine-rank features; and performing fine-rank on the cold start objects in the fine-rank pool based on the above-mentioned plurality of fine-rank features. Specifically, the predicted sales volume of the cold start object and other fine-rank features may be input into a gated network and a characterization network together to complete the fine-rank of the cold start object. The specific method for determining a plurality of fine-rank features based on the feature information of the cold start object and the specific fine-rank algorithm may also refer to existing fine-rank algorithms and will not be repeated here.

[0128] In step 640, at least one candidate cold start object is determined from the cold start objects in the fine-rank pool according to the fine-rank result.

[0129] It can be seen from this that in the above embodiment, the cold start object with a predicted sales volume greater than a pre-set sales volume threshold can be used as one path for object recall, and the priority of the object with a high predicted sales volume in the cold start object during rough ranking can be improved through a weighting mechanism. Further, the predicted sales volume of the cold start object is used as one of the fine-rank features to participate in the fine-rank, so as to increase the exclusive recall quota for the cold start object with high sales potential, and further increase the exposure rate of the cold start object with high sales potential, thereby improving the new object introduction mechanism.

[0130] Regarding step 130 above, the candidate non-cold start object mentioned in step 130 refers to the candidate non-cold start object obtained after recall, rough ranking, and fine-ranking of non-cold start objects. The specific methods for recall, rough ranking, and fine-ranking of non-cold start objects may refer to the recall, rough ranking, and fine-ranking methods adopted by existing recommendation systems, that is, they can all be realized through the existing information recommendation process and will not be repeated here.

[0131] In addition, the specific implementation method for selecting the target object for information promotion from the above at least one candidate cold start object and candidate non-cold start object may include: first, performing a mixed ranking on the above at least one candidate cold start object and candidate non-cold start object; then, determining the target object based on the mixed ranking result. In an embodiment of the present disclosure, the above candidate cold start object and the above candidate non-cold start object may be mixed-ranked using the mixed ranking method adopted by existing recommendation platforms to obtain the object that finally needs to be promoted for information.

[0132] Regarding the above-mentioned step 140, the operations for promoting the target object can specifically include: generating a display page for the target object based on some or all of the feature information of the target object; and presenting the display page of the target object to the user. In some embodiments, the display page of the target object may include, for example, presenting some or all of the feature information of the target object to the user, such as presenting the title, relevant description information, associated pictures, and price of the target object, etc.

[0133] In the foregoing information promotion method, through the method of sales quantile sampling and label discretization, the distribution of samples in the training dataset can be optimized, thereby effectively improving the rank correlation and prediction accuracy of the large language model after supervised fine-tuning, and effectively reducing the impact of the long-tail distribution on sales prediction. In addition, in the design of the input prompt words, by jointly learning text features (such as the description information of the object) and image features (such as the pictures associated with the object), the accuracy of the large language model for sales prediction can be improved. Further, the above information promotion method can apply the prediction results of sales to applications such as information recommendation or operation support, etc., to enhance the efficiency of introducing cold-start objects on the e-commerce platform.

[0134] Specifically, through the large language model, rich object features can be extracted from unstructured text data, effectively improving the object screening and sales prediction capabilities. Especially in the case of scarce data, the prediction effect of the sales prediction model can be significantly improved. Especially in the new product introduction stage, by improving the sales prediction ability of new products, the effective allocation of operation resources can be guided. Based on the existing information recommendation process, an exclusive recall quota for cold-start objects can be increased, so as to help cold-start objects quickly break through the cold-start period and improve the performance of new products on the e-commerce platform.

[0135] In addition, through accurate sales prediction and ranking, the information promotion for the target object is more in line with the needs of in-site users, improving the efficiency and accuracy of introducing objects across the network. Especially in multi-channel promotion, the sales conversion can be significantly improved, driving the growth of objects.

[0136] Corresponding to the above information promotion method, some embodiments of the present disclosure also disclose an information promotion device. Figure 7 Shows the internal structure of the information promotion device described in the embodiments of the present disclosure. As Figure 7 shown, the above information promotion device may include the following multiple modules:

[0137] A sales prediction module 710, configured to respectively determine the predicted sales volume of each cold-start object in the cold-start object pool based on a sales prediction model; wherein, the sales prediction model is a large language model after supervised fine-tuning;

[0138] A candidate cold start object determination module 720, configured to determine at least one candidate cold start object based on the predicted sales volumes of each cold start object;

[0139] A target object determination module 730, configured to select a target object for information promotion from at least one candidate cold start object and candidate non-cold start objects; and

[0140] An information promotion module 740, configured to promote the target object.

[0141] It should be noted that the implementation methods of each module in the above device and the specific technical effects that can be achieved can refer to the implementation methods of each step in the foregoing embodiments, and will not be repeated here. In addition, the methods for multi-round preference alignment fine-tuning and supervised fine-tuning of the large language model can also refer to the specific methods described in the previous embodiments, and will not be repeated here.

[0142] Based on the same inventive concept, corresponding to the method of any of the above embodiments, the present disclosure also provides an electronic device, including: a memory, a processor, and a computer program stored on the memory and executable on the processor, where the processor, when executing the program, implements the information promotion method described in any of the above embodiments.

[0143] Figure 8 FIG. shows a schematic hardware structure diagram of a more specific electronic device provided in this embodiment. The device may include: a processor 2010, a memory 2020, an input / output interface 2030, a communication interface 2040, and a bus 2050. Among them, the processor 2010, the memory 2020, the input / output interface 2030, and the communication interface 2040 are communicatively connected to each other inside the device through the bus 2050.

[0144] The processor 2010 may be implemented in a general-purpose CPU (Central Processing Unit), a microprocessor, an application-specific integrated circuit (ASIC), or one or more integrated circuits, etc., and is configured to execute relevant programs to implement the technical solutions provided in the embodiments of this specification.

[0145] The memory 2020 can be implemented in the form of ROM (Read Only Memory), RAM (Random Access Memory), static storage devices, dynamic storage devices, etc. The memory 2020 can store an operating system and other application programs. When implementing the technical solutions provided in the embodiments of this specification through software or firmware, the relevant program codes are stored in the memory 2020 and called and executed by the processor 2010.

[0146] The input / output interface 2030 is used to connect input / output devices to achieve information input and output. Among them, the input / output devices can be configured as components in the device or externally connected to the device to provide corresponding functions. The input devices can include microphones, various sensors, etc., and the output devices can include displays, speakers, vibrators, indicator lights, etc.

[0147] The communication interface 2040 is used to connect a communication module (not shown in the figure) to achieve communication interaction between this device and other devices. The communication module can achieve communication through wired means (such as USB, network cable, etc.) or through wireless means (such as mobile network, WIFI, Bluetooth, etc.).

[0148] The bus 2050 includes a path for transmitting information between various components of the device (such as the processor 2010, the memory 2020, the input / output interface 2030, and the communication interface 2040).

[0149] It should be noted that although the above device only shows the processor 2010, the memory 2020, the input / output interface 2030, the communication interface 2040, and the bus 2050, in the specific implementation process, the device may also include other components necessary for normal operation. In addition, those skilled in the art can understand that the above device may also only include the components necessary to implement the solutions of the embodiments of this specification, and do not necessarily include all the components shown in the figure.

[0150] The electronic device in the above embodiment is used to implement the corresponding information promotion method in any of the foregoing embodiments, and has the beneficial effects of the corresponding method embodiments, which will not be elaborated here.

[0151] Based on the same inventive concept, corresponding to the method in any of the above embodiments, the present disclosure also provides a non-transitory computer-readable storage medium storing computer instructions for causing the computer to execute the information promotion method as described in any of the foregoing embodiments.

[0152] The computer-readable media of this embodiment include permanent and non-permanent, removable and non-removable media, and information storage can be implemented by any method or technology. The information can be computer-readable instructions, data structures, program modules, or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, compact disc read-only memory (CD-ROM), digital versatile disc (DVD) or other optical storage, magnetic cassette tapes, magnetic tape magnetic disk storage or other magnetic storage devices, or any other non-transmission media that can be used to store information accessible by a computing device.

[0153] The computer instructions stored in the storage medium of the above embodiment are used to cause the computer to execute the task processing method described in any of the above embodiments, and have the beneficial effects of the corresponding method embodiments, which will not be elaborated here.

[0154] Those of ordinary skill in the art should understand that: the discussion of any of the above embodiments is only exemplary, and is not intended to imply that the scope of the present disclosure (including the claims) is limited to these examples; under the concept of the present disclosure, the technical features in the above embodiments or different embodiments can also be combined, the steps can be implemented in any order, and there are many other variations in different aspects of the embodiments of the present disclosure as described above, and they are not provided in detail for the sake of brevity.

[0155] In addition, for simplicity of description and discussion, and in order not to make the embodiments of the present disclosure difficult to understand, the well-known power / ground connections to integrated circuit (IC) chips and other components may or may not be shown in the provided drawings. In addition, the devices may be shown in block diagram form to avoid making the embodiments of the present disclosure difficult to understand, and this also takes into account the fact that the details of the implementation of these block diagram devices are highly dependent on the platform on which the embodiments of the present disclosure will be implemented (i.e., these details should be completely within the understanding of those skilled in the art). In the case where specific details (such as circuits) are set forth to describe the exemplary embodiments of the present disclosure, it will be apparent to those skilled in the art that the embodiments of the present disclosure can be implemented without these specific details or with variations of these specific details. Therefore, these descriptions should be considered illustrative rather than restrictive.

[0156] Although the present disclosure has been described in connection with specific embodiments thereof, many alternatives, modifications, and variations of these embodiments will be apparent to those of ordinary skill in the art in light of the foregoing description. For example, other memory architectures (e.g., dynamic RAM (DRAM)) may be used with the embodiments discussed.

[0157] Embodiments of the present disclosure also provide a computer program product, including computer program instructions which, when run on a computer, cause the computer to execute the above-described information promotion method.

[0158] Embodiments of the present disclosure are intended to cover all such alternatives, modifications, and variations that fall within the broad scope of the appended claims. Accordingly, any omissions, modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the embodiments of the present disclosure shall be included within the protection scope of the present disclosure.

Claims

1. An information promotion method, comprising: Determine the predicted sales volume of each cold start object in the cold start object pool based on the sales volume prediction model; wherein the sales volume prediction model is a large language model that has undergone supervised fine-tuning; Determine at least one candidate cold start object based on the predicted sales volume of each cold start object; Selecting a target object for information promotion from the at least one candidate cold-start object and the candidate non-cold-start object; and Promote the target audience.

2. The method according to claim 1, further comprising: Constructing a training data set; wherein the training data set includes: a plurality of training samples and discrete sales labels corresponding to the training samples; Generating sales prediction prompt words for each training sample in the training data set; Inputting the sales forecast prompt words into a large language model; Acquire a plurality of predicted sales volumes respectively corresponding to the plurality of training samples output by the large language model; Determining, for each training sample, the loss between the predicted sales volume corresponding to the training sample and the discrete sales volume label corresponding to the training sample; and The large language model is fine-tuned in a supervised manner based on the loss to obtain the large language model that has undergone supervised fine-tuning.

3. The method according to claim 2, wherein: The constructing of the training data set comprises: Label the corresponding sales values ​​of multiple samples in the preset sample data set respectively; For samples with sales volume label value of zero, the discrete sales volume label corresponding to the sample is set to zero; For samples whose sales volume annotation values ​​are not zero, the samples are divided into N sample groups according to the preset number of groups N and the N quantiles to which the sales volume annotation values ​​corresponding to the samples belong, and the same discrete sales volume labels are set for the samples in each sample group; wherein, the sample group to which the sample with a larger sales volume annotation value belongs has a larger discrete sales volume label set; N is a positive integer; and For each discrete sales volume label, a predetermined number of samples are extracted from samples corresponding to the discrete sales volume label as the training samples.

4. The method according to claim 2, wherein: Generating sales prediction prompt words for each training sample in the training data set includes: The sales forecast prompt word is generated based on a preset sales forecast prompt word template and feature information of a training sample; wherein, The characteristic information of the training sample includes: basic information of the object, image information of the object, minimum inventory unit (SKU) configuration information of the object, price information of the object, and seller information of the object, or any combination thereof; The preset sales forecast prompt word template includes: a role description part, a background information part, an input data description part, an output requirement part and a special description part.

5. The method according to claim 1, wherein: Determining at least one candidate cold start object based on the predicted sales volume of each cold start object includes: Adding at least one cold start object whose predicted sales volume is greater than a preset sales volume threshold into a cold start object recall pool; performing a rough sorting of the cold start objects in the cold start object recall pool based on the feature information of the cold start objects, and selecting a predetermined number of cold start objects from the cold start objects according to the rough sorting result to add to the fine sorting pool; Performing fine sorting on the cold start objects in the fine sorting pool based on the feature information of the cold start objects; and The at least one candidate cold start object is determined from the cold start objects in the refined ranking pool according to the refined ranking result.

6. The method according to claim 1, wherein: Determining at least one candidate cold start object based on the predicted sales volume of each cold start object includes: Adding at least one cold start object whose predicted sales volume is greater than a preset sales volume threshold into a cold start object recall pool; Roughly sorting the cold start objects in the cold start object recall pool based on the feature information of the cold start objects and the predicted sales volume of the cold start objects, and selecting a predetermined number of cold start objects from the cold start objects according to the rough sorting result to add to the fine sorting pool; Performing fine sorting on the cold start objects in the fine sorting pool based on the feature information of the cold start objects; and The at least one candidate cold start object is determined from the cold start objects in the refined ranking pool according to the refined ranking result.

7. The method according to claim 1, wherein: Determining at least one candidate cold start object based on the predicted sales volume of each cold start object includes: Adding at least one cold start object whose predicted sales volume is greater than a preset sales volume threshold into a cold start object recall pool; Roughly sorting the cold start objects in the cold start object recall pool based on the feature information of the cold start objects and the predicted sales volume of the cold start objects, and selecting a predetermined number of cold start objects from the cold start objects according to the rough sorting result to add to the fine sorting pool; Performing fine sorting on the cold start objects in the fine sorting pool based on the feature information of the cold start objects and the predicted sales volume of the cold start objects; and The at least one candidate cold start object is determined from the cold start objects in the refined ranking pool according to the refined ranking result.

8. The method according to claim 6 or 7, wherein: Roughly sorting the cold start objects in the cold start object recall pool based on the feature information of the cold start objects and the predicted sales volume of the cold start objects includes: The following operations are performed for each cold start object in the cold start object recall pool: determining a rough ranking feature index of the cold start object based on feature information of the cold start object; performing weighted summation on the predicted sales volume of the cold start object and the feature index of the cold start object to obtain a rough ranking comprehensive index of the cold start object; wherein the sum of the weight coefficient of the predicted sales volume of the cold start object and the weight coefficient of the rough ranking feature index of the cold start object is equal to 1; and The cold start objects in the cold start object recall pool are roughly sorted based on the rough sorting comprehensive index of the cold start objects.

9. The method according to claim 7, wherein: Finely sorting the cold start objects in the fine sorting pool based on the characteristic information of the cold start objects and the predicted sales volume of the cold start objects includes: Determine a plurality of refined ranking features based on the feature information of the cold start object and the predicted sales volume of the cold start object; wherein the predicted sales volume of the cold start object is used as one of the plurality of refined ranking features; and The cold start objects in the fine ranking pool are finely ranked based on the multiple fine ranking features.

10. The method according to claim 1, wherein: Determining a target object for information promotion from the at least one candidate cold start object and the determined candidate non-cold start object includes: performing mixed sorting on the at least one candidate cold-start object and the candidate non-cold-start object; and The target object is determined based on the mixed sorting result.

11. An information promotion device, comprising: A sales prediction module, used to determine the predicted sales of each cold start object in the cold start object pool based on a sales prediction model; wherein the sales prediction model is a large language model that has undergone supervised fine-tuning; A candidate cold start object determination module, configured to determine at least one candidate cold start object based on the predicted sales volume of each cold start object; a target object determination module, configured to select a target object for information promotion from the at least one candidate cold start object and the candidate non-cold start object; and The information promotion module is used to promote the target object.

12. An electronic device comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the information promotion method according to any one of claims 1 to 10 when executing the computer program.

13. A non-transitory computer-readable storage medium storing computer instructions, wherein the computer instructions are used to enable a computer to execute the information promotion method according to any one of claims 1 to 10.

14. A computer program product, comprising computer program instructions, which, when executed on a computer, enable the computer to execute the information promotion method according to any one of claims 1 to 10.

Images