Information flow recommendation method and device based on large language model and electronic equipment

Abstract

The invention provides an information flow recommendation method and device based on a large language model and electronic equipment, and relates to the technical field of artificial intelligence, in particular to the technical fields of large language models, model fine tuning, deep learning, intelligent search and the like. According to the specific implementation scheme, after an information flow obtaining request is obtained, according to the number of interaction times of a target account in the request for pushed information flow, under the condition that the number of interaction times is smaller than or equal to a preset number threshold value, user interest point information is obtained based on a large language model; and performing information flow recommendation processing according to a target information flow matched with the user interest point information. Therefore, under the condition of less interaction between the target account and the pushed information flow, the interest point information of the user can be accurately obtained based on the large language model, and information flow recommendation is realized according to the target information flow matched with the interest point information of the user.

Description

Technical Field

[0001] The present disclosure relates to the field of artificial intelligence technology, and in particular to technical fields such as large language models, model fine-tuning, deep learning, and intelligent search, and in particular to an information flow recommendation method, device, and electronic device based on a large language model. Background Art

[0002] At present, in the information flow recommendation scenario, the user's interest point information is usually determined based on the user's historical operation behavior on the pushed information flow, and the information flow recommendation is performed based on the user's interest point information. However, the above method is not suitable for the scenario where the user has no historical operation behavior, or has few historical operation behaviors. Therefore, when the user has no historical operation behavior or has few historical operation behaviors on the pushed information flow, how to push the information flow is a problem that needs to be solved urgently. Summary of the invention

[0003] The present disclosure provides a method, device and electronic device for recommending information flow based on a large language model.

[0004] According to one aspect of the present disclosure, there is provided an information flow recommendation method based on a large language model, the method comprising: obtaining an information flow acquisition request, wherein the information flow acquisition request comprises: a target account; obtaining the number of interactions of the target account with respect to a pushed information flow; when the number of interactions is less than or equal to a preset number threshold, obtaining user interest point information, wherein the user interest point information is obtained based on a target large language model; and performing information flow recommendation processing according to a target information flow matching the user interest point information.

[0005] According to another aspect of the present disclosure, there is provided an information flow recommendation device based on a large language model, the device comprising: a first acquisition module, used to obtain an information flow acquisition request, wherein the information flow acquisition request comprises: a target account; a second acquisition module, used to obtain the number of interactions of the target account for the pushed information flow; a third acquisition module, used to obtain user interest point information when the number of interactions is less than or equal to a preset number threshold, wherein the user interest point information is obtained based on a target large language model; a recommendation processing module, used to perform information flow recommendation processing according to a target information flow matching the user interest point information.

[0006] According to another aspect of the present disclosure, an electronic device is provided, comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the information flow recommendation method based on the large language model proposed above in the present disclosure.

[0007] According to another aspect of the present disclosure, a non-transitory computer-readable storage medium storing computer instructions is provided, wherein the computer instructions are used to enable a computer to execute the information flow recommendation method based on a large language model proposed above in the present disclosure.

[0008] According to another aspect of the present disclosure, a computer program product is provided, including a computer program, which, when executed by a processor, implements the steps of the information flow recommendation method based on a large language model proposed above in the present disclosure.

[0009] It should be understood that the content described in this section is not intended to identify the key or important features of the embodiments of the present disclosure, nor is it intended to limit the scope of the present disclosure. Other features of the present disclosure will become easily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] The accompanying drawings are used to better understand the present solution and do not constitute a limitation of the present disclosure.

[0011] Figure 1 is a schematic diagram according to a first embodiment of the present disclosure;

[0012] Figure 2 is a schematic diagram according to a second embodiment of the present disclosure;

[0013] Figure 3 is a schematic diagram according to a third embodiment of the present disclosure;

[0014] Figure 4 It is a schematic diagram of the recommended framework of information flow;

[0015] Figure 5 is a schematic diagram according to a fourth embodiment of the present disclosure;

[0016] Figure 6 It is a block diagram of an electronic device used to implement the information flow recommendation method based on a large language model according to an embodiment of the present disclosure. DETAILED DESCRIPTION

[0017] The following is a description of exemplary embodiments of the present disclosure in conjunction with the accompanying drawings, including various details of the embodiments of the present disclosure to facilitate understanding, which should be considered as merely exemplary. Therefore, it should be recognized by those of ordinary skill in the art that various changes and modifications may be made to the embodiments described herein without departing from the scope and spirit of the present disclosure. Similarly, for the sake of clarity and conciseness, descriptions of well-known functions and structures are omitted in the following description.

[0018] At present, in the information flow recommendation scenario, the user's interest point information is usually determined based on the user's historical operation behavior on the pushed information flow, and the information flow recommendation is performed based on the user's interest point information. However, the above method is not suitable for the scenario where the user has no historical operation behavior, or has few historical operation behaviors. Therefore, when the user has no historical operation behavior or has few historical operation behaviors on the pushed information flow, how to push the information flow is a problem that needs to be solved urgently.

[0019] In view of the above problems, the present disclosure proposes an information flow recommendation method, device and electronic device based on a large language model.

[0020] Figure 1 It is a schematic diagram according to the first embodiment of the present disclosure. It should be noted that the information flow recommendation method based on the large language model of the embodiment of the present disclosure can be applied to the information flow recommendation device based on the large language model. The device can be configured in an electronic device so that the electronic device can perform the information flow recommendation function.

[0021] The electronic device may be any device with computing capability, for example, the electronic device may be a server, a server set, etc. It should be noted that in the following embodiments, the execution subject is an electronic device as an example for exemplary description.

[0022] like Figure 1 As shown, the information flow recommendation method based on the large language model may include the following steps:

[0023] Step 101, obtaining an information flow acquisition request, wherein the information flow acquisition request includes: a target account.

[0024] The target account refers to the account used by the corresponding user to log in to the target application.

[0025] It should be noted that the client of the target application can be installed on any electronic device, and the corresponding user can log in to the target account of the target application on the client.

[0026] Step 102, obtaining the number of interactions of the target account with respect to the pushed information flow.

[0027] Among them, the pushed information flow refers to the information flow that has been pushed to the target account.

[0028] The number of interactions refers to the total number of times the user of the target account has interacted with the resources in the pushed information flow.

[0029] In this embodiment, the target resources that have interacted with the target account in the pushed information flow can be obtained from the historical behavior data of the target account, and the number of interactions of the target account with respect to the pushed information flow can be determined based on the number of interactions between the target account and the target resource.

[0030] It can be understood that each time the corresponding user performs an interactive operation on the target resource, a corresponding interaction record for the target resource can be added to the historical behavior data, and the number of interactions between the target account and the target resource can be determined based on the number of interaction records of the target resource.

[0031] In this embodiment, the number of interactions between the target account and each target resource may be summed up to obtain the number of interactions of the target account with respect to the pushed information flow.

[0032] Among them, the interactive operations in this embodiment may include but are not limited to operations such as clicking, browsing, liking, commenting and sharing.

[0033] It should be noted that the resources in the pushed information flow can be various types of resources, such as articles, videos, blogs, e-books, sports equipment, online forums, etc., and this embodiment does not make specific limitations here.

[0034] Step 103, when the number of interactions is less than or equal to a preset number threshold, obtaining user interest point information, wherein the user interest point information is obtained based on the target large language model.

[0035] In some exemplary embodiments, the user interest point information is obtained by inputting the user portrait data of the target account into the target large language model.

[0036] It should be noted that the user portrait data of the target account is obtained with the consent of the user corresponding to the target account, and complies with the provisions of relevant laws and regulations and does not violate public order and good customs.

[0037] The preset number threshold is a number threshold pre-set according to actual needs. For example, the preset number threshold may be 5, 6, 8, etc. This embodiment does not specifically limit the value of the preset number threshold.

[0038] In this embodiment, in order to improve the efficiency of obtaining user interest point information, after obtaining the user portrait data of the target account, the user interest point information corresponding to the user portrait data can be obtained from a preset interest point database.

[0039] Specifically, it is possible to query whether the user portrait data exists in the preset point of interest database. If the user portrait data exists in the preset point of interest database, the user point of interest information corresponding to the user portrait data is directly obtained from the point of interest database.

[0040] In addition, when the user portrait data does not exist in the preset interest point database, the user portrait data can also be input into the target large language model to generate the user interest point data through the target large language model.

[0041] The user portrait data refers to the portrait data of the user corresponding to the target account.

[0042] In some embodiments, the user portrait data of the target account may be determined based on the historical behavior data of the target account on the target application and the information provided by the user when registering the target account.

[0043] The user's interest point information refers to interest tags that describe the resources that the user corresponding to the target account is interested in. The interest tags may include but are not limited to: technology, movies, travel, food, health and fitness, fashion and beauty, pets, etc.

[0044] For example, the target account often searches for technology-related words, and accordingly, it can be determined that the interest tag corresponding to the target account is "technology".

[0045] It should be noted that the user's POI information in the POI database is obtained by inputting the user portrait data of the target account into the target large language model. Thus, the user portrait data is analyzed by the target large language model, so as to quickly and accurately obtain the user's POI information.

[0046] In this embodiment, in order to reduce the training cost of the target large language model, the target large language model in this embodiment is fine-tuned on the pre-trained large language model based on the user portrait data and interest point information of the sample account.

[0047] Among them, the sample account refers to the account used when fine-tuning the pre-trained large language model.

[0048] The sample account and the target account in this embodiment may be different accounts registered on the target application. In other embodiments, the sample account may be an account registered on other applications with similar functions to the target application, and the target account may be an account registered on the target application.

[0049] In this embodiment, a possible implementation method of fine-tuning the pre-trained large language model based on the user portrait data and user interest point information of the sample account to obtain a target large language model is as follows: a first prompt word can be generated based on the user portrait data of the sample account, wherein the first prompt word is used to instruct the pre-trained large language model to generate points of interest belonging to a preset set of interest points based on the user portrait data of the sample account, and the first prompt word is input into the pre-trained large language model to obtain an interest point prediction result, and the pre-trained large language model is fine-tuned based on the interest point prediction result and the interest point information of the sample account to obtain a target large language model. As a result, the interest points generated by the trained target large language model are controllable, which can reduce the possibility of the target large language model outputting irrelevant interest points.

[0050] Among them, it can be understood that after obtaining the user interest point information of the sample account, it can be determined whether there are target interest points in the user interest point information of the sample account that are not included in the preset interest point set. If there are target interest points in the user interest point information of the sample account that are not included in the preset interest point set, the target interest points in the user interest point information of the sample account can be deleted.

[0051] In this embodiment, in order to reduce the complexity of model parameter adjustment and training, a low-rank adaptation LoRA (Low-Rank Adaptation) method can be used to fine-tune the pre-trained large language model according to the POI prediction results and the user POI information of the sample account to obtain the target large language model. Due to the LoRA method, when fine-tuning the pre-trained large language model, only some parameters in the pre-trained large language model are adjusted, so the number of parameters for model parameter adjustment can be reduced, and the efficiency of obtaining the target large language model can be improved.

[0052] In other exemplary embodiments, the user's historical search terms can also reflect the user's interests. Correspondingly, in order to improve the accuracy of the user's interest point information in the interest point database, the user's interest point information in the interest point database can be obtained by inputting the user portrait data of the target account and the historical search term information of the target account into the target large language model. Thus, by analyzing the user portrait data and historical search term information of the target account through the target large language model, the accuracy of the obtained user interest point information can be further improved.

[0053] Among them, it should be noted that the historical search term information of the target account is also obtained with the user's consent, and complies with the provisions of relevant laws and regulations and does not violate public order and good customs.

[0054] In some examples, in order to further determine the user's recent points of interest information, correspondingly, the historical search term information of the target account within a recent preset time period can be obtained.

[0055] The most recent preset duration refers to the preset duration before the current time.

[0056] The length of the most recent preset duration may be set and adjusted by a technician according to actual application scenarios, and is not specifically limited in this embodiment.

[0057] In some examples, the target large prediction model is obtained by fine-tuning a pre-trained large language model based on the personality data and user interest point information of the sample account, wherein the personalized data includes: user portrait data and historical search term information of the sample account.

[0058] In this embodiment, a possible implementation method of fine-tuning the pre-trained large language model based on the personalized data of the sample account and the user's interest point information to obtain the target large language model is as follows: a second prompt word can be generated based on the personalized data of the sample account, wherein the second prompt word is used to instruct the pre-trained large language model to generate points of interest belonging to a preset interest point set based on the personalized data, and the second prompt word is input into the pre-trained large language model to obtain the interest point prediction result, and the pre-trained large language model is fine-tuned based on the interest point prediction result and the user interest point information of the sample account to obtain the target large language model. As a result, the interest points generated by the trained target large language model are controllable, which can reduce the possibility of the target large language model outputting irrelevant interest points.

[0059] Among them, for a specific description of fine-tuning the pre-trained large language model based on the POI prediction results and the user POI information of the sample account to obtain the target large language model, please refer to the relevant description in the embodiments of the present disclosure, which will not be repeated here.

[0060] It should be noted that the personal data and user interest point information of the sample accounts are obtained with the user's consent, and are in compliance with relevant laws and regulations and do not violate public order and good customs.

[0061] Step 104: perform information flow recommendation processing according to the target information flow that matches the user's interest point information.

[0062] In one embodiment of the present disclosure, a target information stream matching the user's interest point information can be obtained, and the target information stream can be sent to the sender of the information stream acquisition request. Thus, in the case where the user of the target account has less interaction with the pushed information stream, the user's interest point information is determined in combination with the user portrait data of the target account, and information stream recommendation processing is performed based on the user's interest point information, so that the sent information stream matches the interests of the user of the sender, thereby improving the efficiency of the user in obtaining the content of interest.

[0063] The sending end is a terminal device that sends an information flow acquisition request.

[0064] It should be noted that in different application scenarios, the methods for obtaining the target information flow matching the user's point of interest information are different, and an exemplary description is as follows:

[0065] As an example, based on the pre-saved correspondence between the user's interest point information and the information flow, the target information flow matching the user's interest point information can be obtained.

[0066] As another example, target recommendation resources matching the user's point of interest information are obtained from a preset point of interest database, and a target information flow is generated according to the target recommendation resources.

[0067] In this embodiment, one method of obtaining target recommended resources in the preset interest point data is: obtaining candidate recommended resources matching the user's interest point information, and sorting the candidate recommended resources in descending order according to the historical click probability of the candidate recommended resources to obtain a first sorting result; selecting the first k candidate recommended resources from the first sorting result as target recommended resources, where k is a positive integer. Thus, the target recommended resources matching the user's interest point information are quickly obtained from the preset interest point database, and the target information flow matching the user's interest point information is quickly generated based on the target recommended resources.

[0068] The information flow recommendation method based on the large language model of the embodiment of the present disclosure, after obtaining the information flow acquisition request, obtains the user's point of interest information based on the large language model according to the number of interactions of the target account in the request with the pushed information flow, and when the number of interactions is less than or equal to the preset number threshold, and performs information flow recommendation processing according to the target information flow that matches the user's point of interest information. Thus, in the case where there are few interactions between the target account and the pushed information flow, based on the large language model, the user's point of interest information can be accurately obtained, and information flow recommendation can be implemented according to the target information flow that matches the user's point of interest information.

[0069] In order to clearly understand the process of obtaining a target information flow that matches the user's interest point information, the embodiment of the present disclosure also provides an information flow recommendation method based on a large language model.

[0070] Figure 2 is a schematic diagram according to a second embodiment of the present disclosure.

[0071] like Figure 2 As shown, the information flow recommendation method based on the large language model includes the following steps:

[0072] Step 201: Obtain an information flow acquisition request, wherein the information flow acquisition request includes: a target account.

[0073] Step 202, obtaining the number of interactions of the target account with respect to the pushed information flow.

[0074] It should be noted that, for the detailed contents of step 201 to step 202, reference can be made to the relevant description in the above embodiment, and no detailed description will be given here.

[0075] Step 203, when the number of interactions is less than or equal to a preset number threshold, obtaining user interest point information, wherein the user interest point information is obtained based on the target large language model.

[0076] In some exemplary embodiments, the user interest point information is obtained by inputting the user portrait data of the target account into the target large language model.

[0077] It should be noted that, for the detailed content of step 203, reference may be made to the relevant description in the above embodiment, which will not be described in detail here.

[0078] Step 204, when the number of interactions is greater than a preset threshold, obtain the user interest point information corresponding to the target account, wherein the user interest point information corresponding to the target account is determined based on the interest points of the target resource, and the target resource is a resource in the pushed information flow that has interacted with the target account.

[0079] In this embodiment, when the number of interactions is greater than the preset number threshold, the user's point of interest information corresponding to the target account is obtained from the preset point of interest database. Thus, by pre-saving the user's point of interest information of the target account in the point of interest database, the efficiency of obtaining the user's point of interest information can be improved when the number of interactions is greater than the preset number threshold.

[0080] Step 205: Perform information flow recommendation processing according to the target information flow that matches the user's interest point information.

[0081] It should be noted that, for the detailed content of step 205, reference may be made to the relevant description in the above embodiment, which will not be described in detail here.

[0082] The information flow recommendation method based on the large language model of the disclosed embodiment, after obtaining the information flow acquisition request, obtains the user's point of interest information based on the large language model according to the number of interactions of the target account in the request with respect to the pushed information flow, and when the number of interactions is less than or equal to the preset number threshold, directly obtains the user's point of interest information corresponding to the target account when the number of interactions is greater than the preset number threshold, and performs information flow recommendation processing according to the target information flow matching the user's point of interest information. Thus, according to different comparison results, different processing methods are adopted to accurately obtain the user's point of interest information, and based on the obtained user's point of interest information, the information flow push is accurately realized, thereby improving the effect of the information flow recommendation processing.

[0083] Figure 3 is a schematic diagram according to a third embodiment of the present disclosure, Figure 3 The illustrated embodiment may include the following steps:

[0084] Step 301: Obtain an information flow acquisition request, wherein the information flow acquisition request includes: a target account.

[0085] Step 302, obtaining the number of interactions of the target account with respect to the pushed information flow.

[0086] Step 303: when the number of interactions is less than or equal to a preset number threshold, obtain user interest point information, wherein the user interest point information is obtained based on the target large language model.

[0087] It should be noted that, for the specific description of steps 301 to 303, reference may be made to the relevant descriptions in other embodiments, which will not be repeated here.

[0088] Step 304: Obtain target recommended resources that match the user's point of interest information.

[0089] In the disclosed embodiment, the process of the electronic device executing step 303 may be, for example, obtaining candidate recommended resources that match the user's interest point information; and determining the target recommended resources from the candidate recommended resources. Thus, after obtaining the candidate recommended resources that match the user's interest point information, the candidate recommended resources are continuously selected, so that the selected part of the recommended resources can be used to generate the target information stream, and the target information stream is sent to the sender. Thus, while being able to provide the sender with a target information stream that matches its interests, the user's confusion and fatigue when facing a large amount of information can be reduced, thereby improving the experience.

[0090] Among them, the candidate recommended resources and target recommended resources in this embodiment can be various resources that match the user's interest point information, such as articles, videos, blogs, e-books, sports equipment, online forums, etc., and this embodiment does not make specific limitations on this.

[0091] It is understandable that in different application scenarios, the methods for determining the target recommended resource from the candidate recommended resources are different, and an exemplary description is as follows:

[0092] As an example, a predetermined number of candidate recommendation resources may be randomly selected from the candidate recommendation resources as target recommendation resources.

[0093] The preset number is a pre-set number, for example, the preset number may be 100, 200 or 300, etc. This embodiment does not specifically limit the preset number of expulsions.

[0094] As another example, a target recommendation resource that has not been recommended to the target account may be selected from the candidate recommendation resources.

[0095] As another example, the candidate recommended resources are sorted according to the order of their historical click probabilities from large to small to obtain a first sorting result; the first k candidate recommended resources are selected from the first sorting result as target recommended resources, where k is a positive integer. Thus, in combination with the historical click probabilities of the candidate recommended resources, candidate resources with relatively high historical click probabilities are selected from the candidate recommended resources as target recommended resources, thereby further improving the accuracy of the generated target information flow and further improving the matching degree between the generated target information flow and the user interests of the sender.

[0096] Step 305: Generate a target information flow based on the target recommended resources.

[0097] As an example, the target information flow may be directly generated based on the target recommendation resource.

[0098] As another example, the degree of match between the user's point of interest information and the points of interest of the target recommended resources is determined; based on the degree of match, the target recommended resources are sorted in descending order to obtain a second sorting result; and the first N target recommended resources are selected from the second sorting result as the target information flow, where N is a positive integer. Thus, the target recommended resources are sorted in combination with the degree of match between the user's point of interest information and the points of interest of the target recommended resources, and the first few target recommended resources with the highest degree of match with the user's point of interest information are selected from the sorting results as the target information flow, further improving the degree of match between the determined target information flow and the user's point of interest information, thereby improving the efficiency of the sending end user in acquiring resources of the points of interest of interest.

[0099] Step 306: Send the target information flow to the sender of the information flow acquisition request.

[0100] In this embodiment, when the target account has relatively few interactions with the pushed information stream, combined with the target recommendation resources that match the user's interest point information, the target information stream can be accurately generated and sent to the sender of the information stream acquisition request. As a result, when the target account has relatively few interactions with the pushed information stream, the target recommendation resources that match the user's interest point information can be provided to the sender, thereby matching the interests of the user at the sender and improving the efficiency of the user in acquiring content of interest.

[0101] In this embodiment, when the number of interactions of the target account with respect to the pushed information stream is less than or equal to the preset number threshold, the user's point of interest information can be accurately obtained in combination with the user portrait data of the target account, and the target recommended resources matching the user's point of interest information can be obtained, and the target information stream is generated according to the target recommended resources, and the target information stream is sent to the sender of the information stream acquisition request. Thus, when new recommended resources appear, the newly appeared recommended resources can be easily added to the target recommended resources, and a new information stream can be quickly obtained based on the new target recommended resources, without the need to significantly modify the processing logic when the electronic device performs the information stream recommendation function, and can flexibly adapt to changes.

[0102] In order to clearly understand this application, Figure 4 The recommended framework diagram of the information flow is used to exemplarily describe the method of this embodiment. It should be noted that in this embodiment, the target account, the first type of sample account and the second type of sample account are all different accounts registered on the target application.

[0103] exist Figure 4 In the paper, the recommendation framework of information flow includes offline part and online part. The offline part is used to obtain the target large language model suitable for the information flow recommendation scenario and build the interest point database.

[0104] Among them, the online part is used to perform information flow recommendation processing based on the point of interest database.

[0105] Specifically, the following steps may be included:

[0106] Step 401 , constructing training data collected when fine-tuning the pre-trained large language model.

[0107] The training data includes: personalized data of the first type of sample accounts and corresponding user interest point information. In order to make the target large language model obtained after fine-tuning have better generalization ability and accuracy, a preset proportion of personalized data of the second type of sample accounts and corresponding user interest point information can be added to the training data.

[0108] It should be noted that, in this embodiment, the personalized data of the first type of sample accounts includes the user portrait data and historical search term information of the first type of sample accounts, and the personalized data of the second type of sample accounts includes the user portrait data and historical search term information of the second type of sample accounts.

[0109] Among them, the first type of sample account refers to an account whose number of interactions with the pushed information flow is less than or equal to the preset number threshold.

[0110] The second type of sample account refers to an account whose number of interactions with the pushed information flow is greater than the preset number threshold. In other words, the second type of sample account has more interactions with the pushed information.

[0111] The user interest point information of the first type of sample account is obtained by analyzing the historical interaction data of the pushed information flow corresponding to the first type of sample account.

[0112] Among them, the user interest point information of the second type of sample accounts is obtained by analyzing the historical interaction data of the pushed information flow corresponding to the second type of sample accounts.

[0113] The preset ratio refers to the ratio or percentage set when adding additional samples to the training data. For example, the preset ratio can be 5%, 10%, or other values. This means the ratio of the number of second-category sample accounts to be added to the total number of samples in the original training data.

[0114] Step 402: Obtain a preset set of points of interest.

[0115] Step 403, fine-tuning the pre-trained large language model according to the preset interest point set and training data to obtain a target large language model.

[0116] In this embodiment, for each sample in the training data, the user interest point information of the sample in the training data can be compared with a preset interest point information set. If the user interest point information of the sample contains a target interest point that is not included in the preset interest point set, the target interest point in the user interest point information of the sample can be compared.

[0117] The total push amount of resources corresponding to the points of interest in the preset point of interest set is greater than or equal to a preset quantity threshold.

[0118] In this embodiment, according to a preset set of interest points and training data, a pre-trained large language model is fine-tuned to obtain a target large language model. A possible implementation method is: for samples in the training data, a target prompt word can be generated according to the personalized data of the sample, wherein the target prompt word is used to instruct the pre-trained large language model to generate interest points belonging to the preset set of interest points based on the personalized data of the sample, the target prompt word is input into the pre-trained large language model to obtain a user interest point prediction result, and the pre-trained large language model is fine-tuned according to the user interest point information of the sample and the user interest point prediction result to obtain the target large language model.

[0119] Step 404, for any existing account of the target application, when the number of interactions of the existing account with respect to the pushed information flow is less than or equal to a preset number threshold, the user portrait data of the existing account is input into the target large language model to generate corresponding first user interest point information, and a first recommended resource is determined based on the first user interest point information, and the user portrait data and its corresponding first user interest point information and the first recommended resource corresponding to the first user interest point information are added to a preset interest point database.

[0120] Among them, a possible implementation method of determining the first recommended resource based on the first user's interest point information is: obtaining candidate recommended resources that match the first user's interest point information from the resource library, and sorting the candidate recommended resources according to the historical click probability of the candidate recommended resources to obtain a sorting result, and selecting the first m candidate recommended resources from the sorting result as the first recommended resource corresponding to the first user's interest point information, where m is a positive integer.

[0121] Here, m is preset according to actual needs, and this embodiment does not specifically limit the value of m.

[0122] Step 405, when the number of interactions of an existing account with respect to the pushed information stream is greater than a preset threshold number, the second user interest point information of the existing account is determined based on the interest points of the target resource, the target resource being a resource in the pushed information stream that has interacted with the target account, and the second recommended resource is determined based on the second user interest point information, and the existing account and its corresponding second user interest point information and the second target recommended resource of the second user interest point information are added to the preset interest point database.

[0123] It should be noted that the method of determining the first recommended resource based on the second user's interest point information is similar to the above-mentioned method of determining the first recommended resource based on the first user's interest point information, and this embodiment will not be repeated.

[0124] Step 406, obtaining the target account to be processed.

[0125] In some embodiments, an information flow acquisition request may be received, and the account of the user who logs into the target application in the information flow acquisition request may be used as the target account to be processed.

[0126] Step 407, obtaining the number of interactions of the target account with respect to its corresponding pushed information flow.

[0127] Step 408, when the number of interactions is less than or equal to the preset threshold, obtain the third user interest point information matching the user portrait data of the target account and the first target recommended resource corresponding to the third user interest point from the interest point database, and perform information flow recommendation processing based on the third user interest point information and the first target recommended resource.

[0128] Step 409, when the number of interactions is greater than a preset threshold, obtain the fourth user's point of interest information corresponding to the target account and the second target recommended resources corresponding to the fourth user's point of interest information from the point of interest database, and perform information flow recommendation processing based on the fourth user's point of interest information and the second target recommended resources.

[0129] In order to implement the above embodiment, the present disclosure also provides an information flow recommendation device based on a large language model. Figure 5 As shown, Figure 5 is a schematic diagram according to the fourth embodiment of the present disclosure. The information flow recommendation device 50 based on the large language model may include: a first acquisition module 501, a second acquisition module 502, a third acquisition module 503 and a recommendation processing module 504, wherein:

[0130] The first acquisition module 501 is used to acquire an information flow acquisition request, wherein the information flow acquisition request includes: a target account.

[0131] The second acquisition module 502 is used to acquire the number of interactions of the target account with respect to the pushed information flow.

[0132] The third acquisition module 503 is used to acquire user interest point information when the number of interactions is less than or equal to a preset number threshold, wherein the user interest point information is obtained based on the target large language model.

[0133] The recommendation processing module 504 is used to perform information flow recommendation processing according to the target information flow that matches the user's interest point information.

[0134] As a possible implementation of the embodiment of the present disclosure, the device further includes:

[0135] The fourth acquisition module is used to obtain the user interest point information corresponding to the target account when the number of interactions is greater than a preset threshold, wherein the user interest point information corresponding to the target account is determined based on the interest points of the target resource, and the target resource is a resource in the pushed information flow that has interacted with the target account.

[0136] As a possible implementation of the embodiment of the present disclosure, the fourth acquisition module is specifically configured to:

[0137] From the preset points of interest database, obtain the user points of interest information corresponding to the target account.

[0138] As a possible implementation method of the embodiment of the present disclosure, the user interest point information is obtained by inputting the user portrait data and the historical search term information of the target account into the target large language model.

[0139] As a possible implementation method of the embodiment of the present disclosure, the target big prediction model is obtained by fine-tuning the pre-trained big language model based on the personality data and user interest point information of the sample account, wherein the personalized data includes: user portrait data and historical search term information of the sample account.

[0140] As a possible implementation of the embodiment of the present disclosure, the recommendation processing module 504 includes:

[0141] An acquisition unit, used to acquire target recommended resources matching the user's interest point information;

[0142] A generating unit, used for generating a target information flow according to target recommendation resources;

[0143] The sending unit is used to send the target information flow to the sending end of the information flow acquisition request.

[0144] As a possible implementation of the embodiment of the present disclosure, the acquisition unit is specifically configured to:

[0145] Obtain candidate recommended resources that match the user's point of interest information;

[0146] Determine the target recommended resource from the candidate recommended resources.

[0147] As a possible implementation of the embodiment of the present disclosure, the acquisition unit is specifically configured to:

[0148] Sorting the candidate recommended resources according to the order of their historical click probabilities from large to small to obtain a first sorting result;

[0149] The first k candidate recommended resources are selected from the first sorting result as target recommended resources, where k is a positive integer.

[0150] As a possible implementation of the embodiment of the present disclosure, the generating unit is specifically configured to:

[0151] Determine the matching degree between the user's point of interest information and the points of interest of the target recommended resource;

[0152] According to the matching degree, the target recommended resources are sorted in descending order to obtain a second sorting result;

[0153] A target information flow is generated according to the target recommended resources ranked in the top N positions in the second sorting result, where N is a positive integer.

[0154] It should be noted that the aforementioned explanation of the information flow recommendation method based on the large language model is also applicable to the information flow recommendation device based on the large language model of this embodiment, and will not be repeated here.

[0155] The information flow recommendation device based on the large language model of the embodiment of the present disclosure, after obtaining the information flow acquisition request, obtains the user's point of interest information based on the large language model according to the number of interactions of the target account in the request with the pushed information flow, and when the number of interactions is less than or equal to the preset number threshold, and performs information flow recommendation processing according to the target information flow matching the user's point of interest information. Thus, in the case where there are few interactions between the target account and the pushed information flow, based on the large language model, the user's point of interest information can be accurately obtained, and information flow recommendation can be implemented according to the target information flow matching the user's point of interest information.

[0156] In the technical solution of the present disclosure, the collection, storage, use, processing, transmission, provision and disclosure of user personal information are all carried out with the user's consent, comply with the relevant laws and regulations, and do not violate public order and good morals.

[0157] According to an embodiment of the present disclosure, the present disclosure also provides an electronic device, a readable storage medium and a computer program product.

[0158] Figure 6 A schematic block diagram of an example electronic device 600 that can be used to implement an embodiment of the present disclosure is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital processing, cellular phones, smart phones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the present disclosure described and / or required herein.

[0159] like Figure 6 As shown, the device 600 includes a computing unit 601, which can perform various appropriate actions and processes according to a computer program stored in a read-only memory (ROM) 602 or a computer program loaded from a storage unit 608 into a random access memory (RAM) 603. In the RAM 603, various programs and data required for the operation of the device 600 can also be stored. The computing unit 601, the ROM 602, and the RAM 603 are connected to each other via a bus 604. An input / output (I / O) interface 605 is also connected to the bus 604.

[0160] A number of components in the device 600 are connected to the I / O interface 605, including: an input unit 606, such as a keyboard, a mouse, etc.; an output unit 607, such as various types of displays, speakers, etc.; a storage unit 608, such as a disk, an optical disk, etc.; and a communication unit 609, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 609 allows the device 600 to exchange information / data with other devices through a computer network such as the Internet and / or various telecommunication networks.

[0161] The computing unit 601 may be a variety of general and / or special processing components with processing and computing capabilities. Some examples of the computing unit 601 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various dedicated artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, digital signal processors (DSPs), and any appropriate processors, controllers, microcontrollers, etc. The computing unit 601 performs the various methods and processes described above, such as the information flow recommendation method based on the large language model. For example, in some embodiments, the information flow recommendation method based on the large language model may be implemented as a computer software program, which is tangibly contained in a machine-readable medium, such as a storage unit 608. In some embodiments, part or all of the computer program may be loaded and / or installed on the device 600 via the ROM 602 and / or the communication unit 609. When the computer program is loaded into the RAM 603 and executed by the computing unit 601, one or more steps of the information flow recommendation method based on the large language model described above may be performed. Alternatively, in other embodiments, the computing unit 601 may be configured to execute the information flow recommendation method based on the large language model in any other appropriate manner (for example, by means of firmware).

[0162] Various implementations of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chips (SOCs), load programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various implementations can include: being implemented in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which can be a special purpose or general purpose programmable processor that can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.

[0163] The program code for implementing the method of the present disclosure may be written in any combination of one or more programming languages. These program codes may be provided to a processor or controller of a general-purpose computer, a special-purpose computer, or other programmable data processing device, so that the program code, when executed by the processor or controller, enables the functions / operations specified in the flow chart and / or block diagram to be implemented. The program code may be executed entirely on the machine, partially on the machine, partially on the machine and partially on a remote machine as a stand-alone software package, or entirely on a remote machine or server.

[0164] In the context of the present disclosure, a machine-readable medium may be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, device, or equipment. A machine-readable medium may be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium may include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, device, or equipment, or any suitable combination of the foregoing. A more specific example of a machine-readable storage medium may include an electrical connection based on one or more lines, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0165] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user can provide input to the computer. Other types of devices can also be used to provide interaction with the user; for example, the feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including acoustic input, voice input, or tactile input).

[0166] The systems and techniques described herein may be implemented in a computing system that includes back-end components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes front-end components (e.g., a user computer with a graphical user interface or a web browser through which a user can interact with implementations of the systems and techniques described herein), or a computing system that includes any combination of such back-end components, middleware components, or front-end components. The components of the system may be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: a local area network (LAN), a wide area network (WAN), and the Internet.

[0167] A computer system may include a client and a server. The client and the server are generally remote from each other and usually interact through a communication network. The relationship of client and server is generated by computer programs running on respective computers and having a client-server relationship with each other. The server may be a cloud server, a server of a distributed system, or a server combined with a blockchain.

[0168] It should be understood that the various forms of processes shown above can be used to reorder, add or delete steps. For example, the steps recorded in this disclosure can be executed in parallel, sequentially or in different orders, as long as the desired results of the technical solutions disclosed in this disclosure can be achieved, and this document does not limit this.

[0169] The above specific implementations do not constitute a limitation on the protection scope of the present disclosure. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions can be made according to design requirements and other factors. Any modification, equivalent substitution and improvement made within the spirit and principle of the present disclosure shall be included in the protection scope of the present disclosure.

Claims

1. An information flow recommendation method based on a large language model, comprising: Obtaining an information flow acquisition request, wherein the information flow acquisition request includes: a target account; Obtain the number of interactions of the target account with respect to the pushed information flow; When the number of interactions is less than or equal to a preset number threshold, obtaining user interest point information, wherein the user interest point information is obtained based on a target large language model; Information flow recommendation processing is performed according to the target information flow that matches the user's interest point information.

2. The method according to claim 1, wherein: The method further comprises: When the number of interactions is greater than a preset threshold, the user interest point information corresponding to the target account is obtained, wherein the user interest point information corresponding to the target account is determined based on the interest points of the target resource, and the target resource is a resource in the pushed information flow that has interacted with the target account.

3. The method according to claim 1, wherein: The obtaining the user interest point information corresponding to the target account includes: The user's point of interest information corresponding to the target account is obtained from a preset point of interest database.

4. The method according to any one of claims 1 to 3, wherein: The information flow recommendation process according to the target information flow matching the user interest point information includes: Obtain target recommended resources that match the user's point of interest information; Generating the target information flow according to the target recommended resources; The target information flow is sent to a sender of the information flow acquisition request.

5. The method according to claim 4, wherein: The acquiring of target recommended resources matching the user's point of interest information includes: Acquire candidate recommended resources matching the user's point of interest information; The target recommended resource is determined from the candidate recommended resources.

6. The method according to claim 5, wherein: The step of determining the target recommended resource from the candidate recommended resources includes: Sorting the candidate recommended resources according to the order of historical click probabilities of the candidate recommended resources from large to small to obtain a first sorting result; The first k candidate recommended resources are selected from the first sorting result as the target recommended resources, where k is a positive integer.

7. The method according to claim 4, wherein: The step of generating the target information flow according to the target recommendation resource includes: Determine the matching degree between the user's interest point information and the interest points of the target recommended resource; According to the matching degree, the target recommended resources are sorted in descending order to obtain a second sorting result; The target information flow is generated according to the target recommended resources ranked in the top N positions in the second sorting result, where N is a positive integer.

8. An information flow recommendation device, comprising: A first acquisition module is used to acquire an information flow acquisition request, wherein the information flow acquisition request includes: a target account; A second acquisition module is used to acquire the number of interactions of the target account with respect to the pushed information flow; A third acquisition module is used to acquire user interest point information when the number of interactions is less than or equal to a preset number threshold, wherein the user interest point information is obtained based on a target large language model; The recommendation processing module is used to perform information flow recommendation processing according to the target information flow matching the user's interest point information.

9. The device according to claim 8, wherein: The device also includes: The fourth acquisition module is used to obtain the user interest point information corresponding to the target account when the number of interactions is greater than a preset threshold, wherein the user interest point information corresponding to the target account is determined based on the interest points of the target resource, and the target resource is the resource in the pushed information flow that has interacted with the target account.

10. The device according to claim 9, wherein: The fourth acquisition module is specifically used for: The user's point of interest information corresponding to the target account is obtained from a preset point of interest database.

11. The device according to any one of claims 8 to 10, wherein: The recommendation processing module includes: An acquisition unit, configured to acquire target recommended resources matching the user's point of interest information; A generating unit, configured to generate the target information flow according to the target recommended resources; A sending unit is used to send the target information flow to a sending end of the information flow acquisition request.

12. The device according to claim 11, wherein The acquisition unit is specifically used for: Acquire candidate recommended resources matching the user's point of interest information; The target recommended resource is determined from the candidate recommended resources.

13. The device according to claim 12, wherein: The acquisition unit is specifically used for: Sorting the candidate recommended resources according to the order of historical click probabilities of the candidate recommended resources from large to small to obtain a first sorting result; The first k candidate recommended resources are selected from the first sorting result as the target recommended resources, where k is a positive integer.

14. The device according to claim 11, wherein: The generating unit is specifically used for: Determine the matching degree between the user's interest point information and the interest points of the target recommended resource; According to the matching degree, the target recommended resources are sorted in descending order to obtain a second sorting result; The target information flow is generated according to the target recommended resources ranked in the top N positions in the second sorting result, where N is a positive integer.

15. An electronic device, comprising: at least one processor; as well as a memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the method according to any one of claims 1 to 7.

16. A non-transitory computer-readable storage medium storing computer instructions, wherein: The computer instructions are used to make a computer execute the method according to any one of claims 1 to 7.

17. A computer program product comprising a computer program, which, when executed by a processor, implements the method according to any one of claims 1 to 7.

Images