Incremental content recommendation method, device, medium and equipment
Through the incremental content recommendation method, the focus and reasoning model of natural language description are used to automatically identify and deduplicate, and the problems of homogeneity and low matching of content push in the existing technology are solved, achieving personalized and efficient Information push.
Patent Information
- Application Number
- CN202410857338.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-06-28
- Publication Date
- 2024-11-08
- Estimated Expiration
- 2044-06-28
AI Technical Summary
There are problems with homogeneity and low user interest matching in existing content push, and users receive duplicate or invalid information.
Using incremental content recommendation method, by collecting users' natural language description concerns, using preset first and second inference models to summarize and semantic inference, automatically identify and deduplicate, and only push and user attention Point matching information.
It effectively avoids the push of duplicate information, improves the personalization and relevance of content, and ensures that users only receive attention and no duplicate information.
Smart Images

Figure CN118643217B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of content recommendation, and in particular to an incremental content recommendation method, device, medium and equipment. Background Art
[0002] Existing content push includes news push, public account push, live broadcast / bloggers push, etc. The method adopted is that users need to follow the public account / channel / anchor (collectively referred to as channels). After the system background receives the attention information, it will send all future updated content links of the channel to the user.
[0003] In the process of implementing the present invention, the applicant discovered that there are at least the following problems in the prior art:
[0004] The content pushed to users is homogeneous and poorly matched with user interests. Summary of the invention
[0005] The embodiments of the present invention provide an incremental content recommendation method, apparatus, medium and device, which solve the problem that the content pushed to users is homogeneous and has a low matching degree with the user's interests.
[0006] To achieve the above objectives, on the one hand, an embodiment of the present invention provides an incremental content recommendation method, comprising:
[0007] Collecting user's concerns; wherein the user's concerns are described in natural language;
[0008] Obtaining the original content of the latest information of the channel from the preset channel;
[0009] Applying a preset first reasoning model, summarizing the original content according to the focus point, and obtaining summary information corresponding to the focus point;
[0010] In the case where the summary information corresponding to the focus point is not empty, determining a query condition according to the focus point, and searching for historical information that meets the query condition in a preset database according to the query condition; the database is used to store historical information from various channels that matches the user's focus point;
[0011] If no historical information satisfying the query condition is found in the database, updating the summary information corresponding to the focus point to the database according to the query condition, and sending the original content and the summary information to the user; or
[0012] When historical information satisfying the query condition is found in the database, a preset second reasoning model is applied to infer whether the semantics expressed by the summary information corresponding to the focus point and the historical information are consistent;
[0013] If the inference result is that the semantics expressed by the summary information corresponding to the focus point and the historical information are inconsistent, the summary information corresponding to the focus point is updated to the database according to the query condition, and the original content and the summary information are sent to the user.
[0014] In a second aspect, an embodiment of the present invention provides an incremental content recommendation device, comprising:
[0015] A focus collection unit, used to collect user's focus; wherein the user's focus is described in natural language;
[0016] A latest content acquisition unit, used to acquire the original content of the latest information of the channel from a preset channel;
[0017] An information summarization unit, configured to apply a preset first reasoning model to summarize the original content according to the focus point to obtain summary information corresponding to the focus point;
[0018] The first duplicate checking unit is used to determine the query condition according to the focus point when the summary information corresponding to the focus point is not empty, and query the historical information that meets the query condition in a preset database according to the query condition; the database is used to store historical information from various channels that matches the user's focus point; if no historical information that meets the query condition is found in the database, the database update and information push unit is triggered; or, if historical information that meets the query condition is found in the database, the second duplicate checking unit is triggered;
[0019] A second duplication checking unit is used to apply a preset second reasoning model to infer whether the semantics expressed by the summary information corresponding to the focus point and the historical information are consistent; if the reasoning result is that the semantics expressed by the summary information corresponding to the focus point and the historical information are inconsistent, triggering a database update and information push unit;
[0020] The database updating and information pushing unit is used to update the summary information corresponding to the focus point to the database according to the query condition, and send the original content and the summary information to the user.
[0021] In a third aspect, an embodiment of the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements any one of the incremental content recommendation methods described above.
[0022] In a fourth aspect, an embodiment of the present invention provides an electronic device, comprising:
[0023] processor;
[0024] a memory for storing instructions executable by the processor;
[0025] The processor is configured to execute the instructions to implement any one of the incremental content recommendation methods described above.
[0026] The above technical solution has the following beneficial effects: it can automatically identify information content based on user concerns in a natural language description manner, using the preset first reasoning model and the second reasoning model, automatically deduplicate duplicate information, avoid pushing duplicate and invalid information to users, and push the information that users are concerned about to users without duplication. BRIEF DESCRIPTION OF THE DRAWINGS
[0027] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.
[0028] Figure 1 is a flow chart of an incremental content recommendation method according to one embodiment of the present invention;
[0029] Figure 2 It is an architecture diagram of an incremental content recommendation device according to one embodiment of the present invention;
[0030] Figure 3 It is a schematic diagram of a subscription mechanism under the prior art;
[0031] Figure 4 This is another flow chart of an incremental content recommendation method according to one embodiment of the present invention. DETAILED DESCRIPTION
[0032] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0033] First, as Figure 1 As shown, an embodiment of the present invention provides an incremental content recommendation method, comprising:
[0034] Step S10, collecting user's concerns; wherein the user's concerns are described in natural language;
[0035] Step S11, obtaining the original content of the latest information of the channel from the preset channel;
[0036] Step S12, applying a preset first reasoning model, summarizing the original content according to the focus point, and obtaining summary information corresponding to the focus point;
[0037] Step S13, when the summary information corresponding to the focus point is not empty, determining a query condition according to the focus point, and searching for historical information that meets the query condition in a preset database according to the query condition; the database is used to store historical information from various channels that matches the user's focus point;
[0038] Step S14, when no historical information satisfying the query condition is found in the database, the summary information corresponding to the focus point is updated to the database according to the query condition, and the original content and the summary information are sent to the user; or
[0039] Step S15, when historical information that meets the query condition is found in the database, a preset second reasoning model is applied to infer whether the semantics expressed by the summary information corresponding to the focus point and the historical information are consistent;
[0040] Step S16, if the inference result is that the semantics expressed by the summary information corresponding to the focus point and the historical information are inconsistent, the summary information corresponding to the focus point is updated to the database according to the query condition, and the original content and the summary information are sent to the user.
[0041] In some embodiments, the first reasoning model and the second reasoning model may be the same or different; the first reasoning model may be a large model based on the large model theory that can process content including but not limited to video, audio, picture, text, etc.; the second reasoning model may also be a large model based on the large model theory that can process content including but not limited to video, audio, picture, text, etc.; the second reasoning model may also be a large language model; there may be one or more preset channels, and when there are multiple channels, the method steps in the embodiment of the present invention may be performed for each preset channel. The original content of the latest information obtained from the preset channel may be content including but not limited to video, audio, picture, text, etc., and the summary information obtained by the first reasoning model is information composed of natural language; the original content of the latest information may or may not contain the content corresponding to the focus point. When the original content of the latest information contains the content corresponding to the focus point, the first reasoning model infers that the summary information corresponding to the focus point is not empty, otherwise it is inferred that the summary information corresponding to the focus point is empty. When it is inferred that the summary information corresponding to the focus point is empty, the first reasoning model can also directly return the indicative information representing that the summary information corresponding to the focus point is empty. The database records the historical information corresponding to the user's focus that has been pushed to the user. If no historical information matching the focus is found through database query, it means that the latest information corresponding to the summary information corresponding to the current focus is of concern to the user and there is no duplication, so it can be sent to the user. If historical information matching the focus is found, the second reasoning model needs to be used to further determine whether the historical information is consistent with the summary information. If not, it means that the latest information corresponding to the summary information corresponding to the current focus is of concern to the user and there is no duplication, so it can be sent to the user. Otherwise, there is duplication and it cannot be sent to the user.
[0042] The embodiments of the present invention have the following technical effects: according to the user's focus, in a natural language description manner, using a preset first reasoning model and a second reasoning model, information content can be automatically identified, duplicate information can be automatically deduplicated, duplicate and invalid information can be avoided, and information that the user is concerned about can be pushed to the user without duplication.
[0043] Furthermore, the first reasoning model preset by the application summarizes the original content according to the focus point to obtain summary information corresponding to the focus point, including:
[0044] Generate a prompt sentence corresponding to the focus point according to the focus point of the user; the prompt sentence corresponding to the focus point is used to instruct the first reasoning model to try to summarize the content matching the focus point from the original content in a natural language description manner;
[0045] Aggregating the original content and the prompt sentence corresponding to the focus point to obtain a first content to be inferred; wherein the original content includes text, video, picture and / or audio;
[0046] The first content to be inferred is input into the first reasoning model for reasoning. When there is content matching the focus point in the first content to be inferred, the first reasoning model summarizes the content matching the focus point from the first content to be inferred in a natural language description as summary information corresponding to the focus point; when there is no content matching the focus point in the first content to be inferred, the summary information corresponding to the focus point is empty.
[0047] For example, obtain the text description of the user's focus, collect the user's focus, and the user uses natural language to describe, such as "I want to know the discount of a certain brand of shoes." The focus described in the text is adapted into prompt words (prompt sentences), and the focus can be converted into prompt sentences by a pre-trained neural network model, or the corresponding prompt sentences preset for the user's preset focus are stored in a key-value pair manner. When used, the corresponding prompt sentence is retrieved according to the currently acquired focus as a key value, or the focus is adapted into a prompt sentence by manual method; specifically, a prompt word (prompt sentence corresponding to the focus) is generated according to the focus, and the prompt word (prompt sentence corresponding to the focus) is used as one of the inputs of the large model (the first reasoning model). For the above example, the prompt word (prompt sentence corresponding to the focus) can be "Summarize the discount of a certain brand of shoes from the above content, the number of words is within 50-100 words, if it is not mentioned, please answer no", and the prompt word (prompt sentence corresponding to the focus) can be automatically generated or created manually. Determine the information acquisition channel (preset channel), such as news, public accounts, self-media, etc., and even the original information (the original content of the latest information of the preset channel) obtained by means of web crawlers, etc. The information form is not limited to text, pictures, sounds, videos, etc. The channel range can be specified by the user or by the system operator. The original information is obtained from the channel, and the acquisition method is not limited, including subscription, web crawler, API call, etc. While obtaining the original information (the original content of the latest information), the relevant information corresponding to the original information (the original content of the latest information) can also be obtained, such as the original link, etc. Aggregate the original information (the original content of the latest information) and the prompt word (the prompt sentence corresponding to the focus point) to obtain the first content to be inferred. The first inference content can be a combination of text, pictures, sounds (audio), and / or videos, and prompt sentences in natural language text form. When the original information is text, the first inference content is a complete paragraph of text. Use the big model (first inference model) to summarize the original information, and input the first inference content obtained into the big model (first inference model) to perform inference; obtain the inference result. There are only two types of inference results. One is to output the summary information of the original information (the original content of the latest information), and the other is to output no (indicative information when the summary information is empty). When the output is no, the process ends and the latest information is not sent to the user. When the summary information of the output original information is not empty, continue to execute the subsequent steps.
[0048] Furthermore, when the summary information corresponding to the focus point is not empty, determining a query condition according to the focus point, and searching a preset database for historical information that meets the query condition according to the query condition, further includes:
[0049] When the summary information corresponding to the focus point is empty, the original content is discarded and is not sent to the user.
[0050] The summary information is empty, which means that the original content does not contain the content that the user is concerned about, and is not the content that the user is concerned about, so it is not sent to the user.
[0051] Furthermore, the application presets a second reasoning model to reason whether the semantics expressed by the summary information corresponding to the focus point and the historical information are consistent, including:
[0052] The summary information corresponding to the focus point, the historical information and a preset prompt sentence are combined into a second content to be inferred; the preset prompt sentence is used to instruct the second inference model to infer whether the semantics of the summary information corresponding to the focus point and the historical information are consistent in a natural language description;
[0053] The second content to be inferred is input into the second inference model for inference, and an inference result is obtained as to whether the semantics of the summary information corresponding to the focus point and the historical information are consistent; the inference result includes semantic consistency and semantic inconsistency.
[0054] For example, the recorded content (historical information) and the summary information of the original information (the summary information corresponding to the focus point) are combined into a new text (the second content to be inferred). Specifically, the recorded content (historical information) and the summary information of the original information output by the first inference model (the summary information corresponding to the focus point) and a preset prompt word (for example, "Are the meanings of the above two descriptions consistent? Please answer yes or no") are combined into a new text (the second content to be inferred);
[0055] The query results are compared through the second inference model. Specifically, the composed new text (the second content to be inferred) is sent to the large model, such as the second inference model, to perform inference. The purpose of this step is to remove duplicates.
[0056] Second, as Figure 2 As shown, an embodiment of the present invention provides an incremental content recommendation device, comprising:
[0057] The focus collection unit 200 is used to collect the user's focus; wherein the user's focus is described in natural language;
[0058] The latest content acquisition unit 201 is used to acquire the original content of the latest information of the channel from a preset channel;
[0059] An information summarization unit 202 is used to apply a preset first reasoning model to summarize the original content according to the focus point to obtain summary information corresponding to the focus point;
[0060] The first duplicate checking unit 203 is used to determine the query condition according to the focus point when the summary information corresponding to the focus point is not empty, and query the historical information that meets the query condition in a preset database according to the query condition; the database is used to store historical information from various channels that matches the user's focus point; if no historical information that meets the query condition is found in the database, the database update and information push unit is triggered; or, if historical information that meets the query condition is found in the database, the second duplicate checking unit is triggered;
[0061] The second duplicate checking unit 204 is used to apply a preset second reasoning model to infer whether the semantics expressed by the summary information corresponding to the focus point and the historical information are consistent; if the reasoning result is that the semantics expressed by the summary information corresponding to the focus point and the historical information are inconsistent, the database update and information push unit is triggered;
[0062] The database updating and information pushing unit 205 is used to update the summary information corresponding to the focus point to the database according to the query condition, and send the original content and the summary information to the user.
[0063] Furthermore, the information summarization unit 202 includes:
[0064] A focus prompt generating module, used for generating a prompt sentence corresponding to the focus point according to the focus point of the user; the prompt sentence corresponding to the focus point is used to instruct the first reasoning model to try to summarize the content matching the focus point from the original content in a natural language description manner;
[0065] A first inference content aggregation module, used to aggregate the original content and the prompt sentence corresponding to the focus point to obtain a first content to be inferred; wherein the original content includes text, video, picture and / or audio;
[0066] The reasoning and summarization module is used to input the first content to be inferred into the first reasoning model for reasoning. When there is content matching the focus point in the first content to be inferred, the first reasoning model summarizes the content matching the focus point from the first content to be inferred in a natural language description as the summary information corresponding to the focus point; when there is no content matching the focus point in the first content to be inferred, the summary information corresponding to the focus point is empty.
[0067] Furthermore, the first duplication checking unit 203 is also used for:
[0068] When the summary information corresponding to the focus point is empty, the original content is discarded and is not sent to the user.
[0069] Furthermore, the second duplication checking unit 204 comprises:
[0070] A second reasoning content generation module is used to merge the summary information corresponding to the focus point, the historical information and a preset prompt sentence into a second content to be reasoned; the preset prompt sentence is used to instruct the second reasoning model to reason whether the semantics of the summary information corresponding to the focus point and the historical information are consistent in a natural language description;
[0071] The duplicate checking reasoning module is used to input the second content to be inferred into the second reasoning model for reasoning, and obtain the reasoning result of whether the semantics of the summary information corresponding to the focus point and the historical information are consistent; the reasoning result includes semantic consistency and semantic inconsistency.
[0072] The embodiments of the present invention correspond one-to-one to the embodiments of the aforementioned incremental content recommendation method. The embodiments of the present invention can be understood based on the embodiments of the aforementioned incremental content recommendation method, and will not be described in detail here.
[0073] In a third aspect, an embodiment of the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements any one of the incremental content recommendation methods described above.
[0074] The embodiment of the present invention is an embodiment of a computer-readable storage medium storing the aforementioned incremental content recommendation method. The embodiment of the present invention can be understood based on the aforementioned embodiment of the incremental content recommendation method, and will not be described in detail here.
[0075] In a fourth aspect, an embodiment of the present invention provides an electronic device, comprising:
[0076] processor;
[0077] a memory for storing instructions executable by the processor;
[0078] The processor is configured to execute the instructions to implement any one of the incremental content recommendation methods described above.
[0079] The embodiment of the present invention is an embodiment of an electronic device that stores and executes the aforementioned incremental content recommendation method. The embodiment of the present invention can be understood based on the aforementioned embodiment of the incremental content recommendation method, and will not be described in detail here.
[0080] The above technical solution of the embodiment of the present invention is described in detail below in conjunction with specific application examples. For technical details not introduced during the implementation process, please refer to the relevant description in the previous text.
[0081] Under the existing technology, the channel pushes messages to users based on the following Figure 3 The inventor analyzed the existing content push based on the relationship between channels and users and found that the message subscription and publishing mechanism shown in the figure has the following problems: the push granularity is too large: users often only pay attention to part of the content of the channel, but the channel does push all subsequent updates of the channel to the users. After receiving the push, the user needs to browse the complete content before determining whether it is the content they pay attention to; duplicate content cannot be distinguished: users are accustomed to paying attention to multiple channels, and the push of these channels will inevitably have homogeneous content. In this case, the user will receive multiple recommendations of the same content, and the user still needs to browse the content before identifying it.
[0082] The essence of the above technical shortcomings is that they can only push content based on channels, not content. Users can only follow channels, but cannot customize the content they follow.
[0083] To solve the above problems, the embodiments of the present invention provide an incremental content recommendation method and system, which is also a recommendation method and system based on content recognition. Users can customize their focus, and the system will only push news / messages that meet the focus of the user based on the focus of the user. The embodiments of the present invention can be applied to industries such as news, content promotion, advertising, and media, and can also be applied to personalized exclusive services provided to consumers in the field of artificial intelligence.
[0084] The purpose of the embodiment of the present invention is to push only relevant messages of the content that users are interested in. The main technical core achieves this effect based on the content recognition capability of the language model. The core modules of the embodiment of the present invention are a large language model and a database. The large model is used for content recognition, and the database is used to save historical records.
[0085] like Figure 4 As shown, the embodiment of the present invention provides an incremental content recommendation method, which can be briefly summarized into four steps: retrieving information, summarizing content, comparing and removing duplicates, and sending recommendations. The specific steps are as follows:
[0086] Step S401, obtaining a text description of the user's focus, collecting the user's focus, and the user uses natural language to describe, for example, "I want to know the discount of a certain brand of shoes."
[0087] Step S402, adapt the text description into prompt words, specifically, generate prompt words according to the focus point, and the prompt words are used as one of the inputs of the large model. For the above example, the prompt words can be "summarize the discount of a certain brand of shoes from the above content, with a word count of 50-100 words. If it is not mentioned, please answer no", and the prompt words can be automatically generated or manually created.
[0088] Step S403, determine the information acquisition channel, such as news, public accounts, self-media, etc., and even include the original information obtained by web crawlers and other methods. The information form is not limited to text, pictures, sounds, videos, etc. The channel range can be specified by the user or by the system operator. For example, the public accounts, bloggers, self-media and other content publishers that the user follows.
[0089] Step S404, obtaining the original information from the channel, the acquisition method is not limited, including subscription, web crawler, API call, etc. When obtaining the original information, you can also obtain the relevant information corresponding to the original information, such as the original link, etc.
[0090] Step S405, combining the original information and the prompt word, specifically, combining the original information obtained in step S404 and the prompt word obtained in step S402 into a complete text;
[0091] Step S406, use the big model to summarize the original information. Specifically, input the complete text obtained in step S405 into the big model to perform reasoning; obtain the big model reasoning result. There are only two types of big model reasoning results, one is to output the summary information of the original information, and the other is to output no. When the output is no, the process ends. When the output is the summary information of the original information, execute the next step.
[0092] Step S407, database query, specifically, query data in the database, the database stores historical information of user focus points, and the query condition is the limited requirements of the user's focus. In the above example, the query condition can be the brand name, or a combination query of brand name + shoes. The historical data in the database can be entered in advance, or an empty database can be created, and the information in the database can be continuously updated through the recommendation process.
[0093] Step S408, obtain the database query result, the query has only two results, with recorded content or without recorded content (historical information). When there is no recorded content, jump to step S412, if there is recorded content, execute the next step.
[0094] Step S409, combining the summary information of the recorded content (historical information) and the original information (the original content of the latest information) into a new text. Specifically, the recorded content, the summary information of the original information output by the large model in step S406, and a prompt word (for example, "Are the meanings of the above two descriptions consistent? Please answer yes or no") are combined to form a new text;
[0095] Step S410: The large model compares the query results. Specifically, the new text composed in step S409 is sent to the large model for reasoning. The purpose of this step is to remove duplicates.
[0096] Step S411, determine whether the summary information is repeated, specifically, obtain the output result of the large model, if the output result of the large model is yes, the process ends. If the output is no, proceed to the next step.
[0097] Step S412, storing / updating the summary information to the database, specifically, saving the summary information of the original information output by the large model in step S406 in the database. In the above example, the key of the database can be the brand name, and the value of the database is the large model output. Saving the summary information of the original information output by the large model in step S406 in the database includes but is not limited to: if there is already a record for the key, deleting the record, and overwriting the original record of the key with the summary information of the new original information; or if there is already a record for the key, adding the summary information of the original information output by the large model in step S406 as a record for the key, and coexisting with the original record of the key, when querying the database, all records of the key can be returned; or if there is no record for the key, directly adding the summary information of the original information output by the large model in step S406 as a record for the key.
[0098] Step S413, the original information and summary information output by the large model in step S406 and related information of the original information (such as the original link, etc.) are sent to the user account.
[0099] There may be manual participation in the above multiple steps, such as retrieval of information can be done manually or automatically through a program. In the last step, before sending the recommendation, a manual review application can be issued, and the result of the manual review can be obtained. When the manual review result is approved, the large model output and related information (such as the original link, etc.) are sent to the user account. The storage and search functions of the database can be implemented in other ways, such as a file system. But it is preferably through a database. Each time the large model is called, the same large model can be called, or different large models can be called. For example, the large models used in step S406 and step S410 can be the same large model or different large models. In the embodiment of the present invention, the large model called in step S406 does not specifically refer to a large language model, but may also include a multimodal model with visual recognition and speech recognition. Because the large language model can only recognize and analyze text input. The embodiment of the present invention is not limited to the recommendation of text messages. The large model called in step S410 can be a large language model because the input is only text.
[0100] The preferred usage scenario of the embodiment of the present invention is daily recommendation service. The user only needs to define the focus points, and the system will push the latest news at all times.
[0101] The present invention does not involve a search function, that is, after the user defines the focus, the system will only push the latest message from the channel and will not search for historical messages in the channel.
[0102] The embodiments of the present invention have the following technical effects: the information content can be automatically identified according to the user's focus, and relevant information can be pushed to the user. Duplicate information can be automatically removed to avoid pushing duplicate and invalid information to the user.
[0103] Another embodiment is described below:
[0104] Step S501, obtain description and determine channel. The user wants to know the price changes of large model services of various companies. The description is: "Recommend to me the service price change information of all large model companies". The channel is a certain large model public account.
[0105] Step S502, generate a prompt word based on the description, the prompt word is "Please summarize the price change information of large model services in the above content, please return the result in the format of {"Company Name": "cccccc"; "Content": "tttttt"}. If the above content does not involve the price of large model services, return "No"."
[0106] Step S503, the channel receives a latest message. The message content is "Currently, the pricing of ByteDance's Doubao large model is significantly lower than the industry price. According to reports, the price of Doubao's main model in the enterprise market is only 0.0008 yuan / thousand tokens, that is, 0.8 cents can process more than 1,500 Chinese characters, which is 99.3% cheaper than the industry. For reference, the price of the same specification model on the market is generally 0.12 yuan / thousand tokens, which is 150 times the price of the Doubao model."
[0107] Step S504, merge the latest news (original information) of step S503 and the prompt word of step S502 to obtain the merged result, and the merged result is "At present, the pricing of Byte Doubao large models is significantly lower than the industry price. It is reported that the pricing of Doubao's main model in the enterprise market is only 0.0008 yuan / thousand Tokens, that is, 0.8 cents can process more than 1,500 Chinese characters, which is 99.3% cheaper than the industry. For reference, the pricing of the same specification model on the market is generally 0.12 yuan / thousand Tokens, which is 150 times the price of the Doubao model. Please summarize the price change information of the large model service in the above content, and please return the result in the format of {"Company Name": "cccccc"; "Content": "tttttt"}. If the above content does not involve the large model service price, return "No"."
[0108] Step S505, input the merged result into the big model to perform reasoning, and the reasoning result (summary information) is "{"Company Name": "Byte"; "Content": "The price of the Doubao big model in the enterprise market is 0.0008 yuan / thousand tokens, that is, 0.8 cents to process more than 1,500 Chinese characters."}"
[0109] Step S506, query the database for related information. The database has saved the price information of large models of various companies. The query condition is "company name" == "bytes", and the query target is "content".
[0110] Step S507, obtaining the query result, the query result is "taking Doubao universal model pro-32k as an example, calculated according to the "prepaid" model unit price, the monthly price of 10K TPM is 2,000 yuan, and the average price is 0.0046 yuan / thousand Tokens."
[0111] Step S508, merge the query result and the output of step S505 into a prompt word as the input of the large model. "Content 1: The price of Doubao large model in the enterprise market is 0.0008 yuan / thousand tokens, that is, 0.8 cents to process more than 1,500 Chinese characters. Content 2: Taking Doubao universal model pro-32k as an example, according to the "prepaid" model unit price calculation, the monthly price of 10K TPM is 2,000 yuan, and the average price is 0.0046 yuan / thousand tokens. Are the above two paragraphs consistent in expression, please answer yes or no."
[0112] Step S509, input the prompt word obtained in step S508 into the large model reasoning, and the reasoning result is "no".
[0113] Step S510, the large model output of step S505 is stored in the database, the company name is listed as "bytes", and the content is listed as the large model output.
[0114] Step S511, push the large model output and original information of step S505 to the user account, and the process ends.
[0115] In the above process, if the result of step S505 is "No" or the result of step S509 is "Yes", the process ends immediately.
[0116] The following is another embodiment:
[0117] An embodiment of the present invention provides an incremental content recommendation method, comprising:
[0118] S61: Collect user concerns and collection channels;
[0119] S62: Send the raw information collected from the channel to the big model for content summary;
[0120] S63: Query relevant information in a database;
[0121] S64: input the query result obtained in the database and the result of S62 into the large model for content comparison;
[0122] S65: If the content is not duplicated, the content is stored in a database, and the content and the original information are sent to the user.
[0123] Step S62 specifically includes:
[0124] S621: Generate prompt words according to the user's focus;
[0125] S622: combining the prompt word and the original information to form a complete text description;
[0126] S623: Input the text description into the large model to obtain a content summary.
[0127] Step S63 specifically includes:
[0128] S631: Setting database query conditions, the query conditions are part of the output of S623;
[0129] S632: Retrieve results from the database according to the query conditions.
[0130] Step S64 specifically includes:
[0131] S641: Setting a large model prompt word, the prompt word is used to compare the similarity of two paragraphs of text;
[0132] S642: Combine the prompt word of S641, the output of S62 and the query result of S63 into a large model input;
[0133] S643: Input the merged content of S642 into the large model to obtain a comparison result.
[0134] The embodiments of the present invention have the following technical effects: the information content can be automatically identified according to the user's focus, and relevant information can be pushed to the user. Duplicate information can be automatically removed to avoid pushing duplicate and invalid information to the user.
[0135] It should be understood that the specific order or hierarchy of steps in the disclosed process is an example of an exemplary method. Based on design preferences, it should be understood that the specific order or hierarchy of steps in the process can be rearranged without departing from the scope of protection of the present disclosure. The attached method claims present the elements of the various steps in an exemplary order and are not intended to be limited to the specific order or hierarchy described.
[0136] In the above detailed description, various features are grouped together in a single embodiment to simplify the disclosure. This method of disclosure should not be interpreted as reflecting an intention that the embodiments of the claimed subject matter require more features than are clearly stated in each claim. On the contrary, as reflected in the appended claims, the invention is in a state of having less than all the features of the disclosed individual embodiments. Therefore, the appended claims are hereby expressly incorporated into the detailed description, with each claim standing on its own as a separate preferred embodiment of the invention.
[0137] The disclosed embodiments are described above to enable any person skilled in the art to implement or use the present invention. Various modifications of these embodiments are obvious to those skilled in the art, and the general principles defined herein may also be applied to other embodiments without departing from the spirit and scope of the present disclosure. Therefore, the present disclosure is not limited to the embodiments given herein, but is consistent with the broadest scope of the principles and novel features disclosed in this application.
[0138] The above description includes examples of one or more embodiments. Of course, it is impossible to describe all possible combinations of components or methods for the purpose of describing the above embodiments, but it should be recognized by those skilled in the art that the various embodiments may be further combined and arranged. Therefore, the embodiments described herein are intended to cover all such changes, modifications and variations that fall within the scope of protection of the appended claims. In addition, with respect to the term "comprising" used in the specification or claims, the word is covered in a manner similar to the term "including". In addition, any term "or" used in the specification of the claims is intended to mean "non-exclusive or".
[0139] Those skilled in the art may also understand that the various illustrative logical blocks, units, and steps listed in the embodiments of the present invention may be implemented by electronic hardware, computer software, or a combination of the two. In order to clearly demonstrate the interchangeability of hardware and software, the various illustrative components, units, and steps described above have generally described their functions. Whether such functions are implemented by hardware or software depends on the specific application and the design requirements of the entire system. Those skilled in the art may use various methods to implement the described functions for each specific application, but such implementation should not be understood as exceeding the scope of protection of the embodiments of the present invention.
[0140] The various illustrative logic blocks or units described in the embodiments of the present invention can be implemented or operated by a general-purpose processor, a digital signal processor, an application-specific integrated circuit (ASIC), a field programmable gate array or other programmable logic device, a discrete gate or transistor logic, a discrete hardware component, or any combination of the above. The general-purpose processor can be a microprocessor, and optionally, the general-purpose processor can also be any conventional processor, controller, microcontroller or state machine. The processor can also be implemented by a combination of computing devices, such as a digital signal processor and a microprocessor, a plurality of microprocessors, one or more microprocessors combined with a digital signal processor core, or any other similar configuration.
[0141] The steps of the method or algorithm described in the embodiments of the present invention can be directly embedded in hardware, a software module executed by a processor, or a combination of the two. The software module can be stored in a RAM memory, a flash memory, a ROM memory, an EPROM memory, an EEPROM memory, a register, a hard disk, a removable disk, a CD-ROM, or other storage media of any form in the art. Exemplarily, the storage medium can be connected to the processor so that the processor can read information from the storage medium and can write information to the storage medium. Optionally, the storage medium can also be integrated into the processor. The processor and the storage medium can be arranged in an ASIC, and the ASIC can be arranged in a user terminal. Optionally, the processor and the storage medium can also be arranged in different components in the user terminal.
[0142] In one or more exemplary designs, the above functions described in the embodiments of the present invention can be implemented in hardware, software, firmware or any combination of the three. If implemented in software, these functions can be stored on a computer-readable medium, or transmitted in the form of one or more instructions or codes on a computer-readable medium. Computer-readable media include computer storage media and communication media that facilitate the transfer of computer programs from one place to another. The storage medium can be any available medium that can be accessed by any general or special computer. For example, such computer-readable media can include but are not limited to RAM, ROM, EEPROM, CD-ROM or other optical disk storage, disk storage or other magnetic storage devices, or any other medium that can be used to carry or store program codes in the form of instructions or data structures and other forms that can be read by general or special computers, or general or special processors. In addition, any connection can be appropriately defined as a computer-readable medium, for example, if the software is transmitted from a website site, server or other remote resource through a coaxial cable, fiber optic cable, twisted pair, digital subscriber line (DSL) or wirelessly, such as infrared, wireless and microwave, it is also included in the defined computer-readable medium. The disk and disc include compact disk, laser disk, optical disk, DVD, floppy disk and blue-ray disk. Disks usually copy data magnetically, while discs usually copy data optically with lasers. The above combination can also be included in computer readable media.
[0143] The specific implementation methods described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific implementation method of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.
Claims
1. An incremental content recommendation method, characterized in that: include: Collecting user's concerns; wherein the user's concerns are described in natural language; Obtaining the original content of the latest information of the channel from the preset channel; Applying a preset first reasoning model, summarizing the original content according to the focus point, and obtaining summary information corresponding to the focus point; In the case where the summary information corresponding to the focus point is not empty, determining a query condition according to the focus point, and searching for historical information that meets the query condition in a preset database according to the query condition; the database is used to store historical information from various channels that matches the user's focus point; If no historical information satisfying the query condition is found in the database, updating the summary information corresponding to the focus point to the database according to the query condition, and sending the original content and the summary information to the user; or When historical information satisfying the query condition is found in the database, a preset second reasoning model is applied to infer whether the summary information corresponding to the focus point is consistent with the semantics expressed by the historical information; If the inference result is that the semantics expressed by the summary information corresponding to the focus point and the historical information are inconsistent, updating the summary information corresponding to the focus point to the database according to the query condition, and sending the original content and the summary information to the user; The first reasoning model preset by the application summarizes the original content according to the focus point to obtain summary information corresponding to the focus point, including: Generate a prompt sentence corresponding to the focus point according to the focus point of the user; the prompt sentence corresponding to the focus point is used to instruct the first reasoning model to try to summarize the content matching the focus point from the original content in a natural language description manner; Aggregating the original content and the prompt sentence corresponding to the focus point to obtain a first content to be inferred; wherein the original content includes text, video, picture and / or audio; The first content to be inferred is input into the first reasoning model for reasoning. When there is content matching the focus point in the first content to be inferred, the first reasoning model summarizes the content matching the focus point from the first content to be inferred in a natural language description as summary information corresponding to the focus point; when there is no content matching the focus point in the first content to be inferred, the summary information corresponding to the focus point is empty.
2. The incremental content recommendation method according to claim 1, characterized in that: In the case where the summary information corresponding to the focus point is not empty, determining a query condition according to the focus point, and searching a preset database for historical information that meets the query condition according to the query condition, further comprising: When the summary information corresponding to the focus point is empty, the original content is discarded and is not sent to the user.
3. The incremental content recommendation method according to claim 1, characterized in that: The application presets a second reasoning model to reason whether the summary information corresponding to the focus point and the semantics expressed by the historical information are consistent, including: The summary information corresponding to the focus point, the historical information and a preset prompt sentence are combined into a second content to be inferred; the preset prompt sentence is used to instruct the second inference model to infer whether the semantics of the summary information corresponding to the focus point and the historical information are consistent in a natural language description; The second content to be inferred is input into the second inference model for inference, and an inference result is obtained as to whether the semantics of the summary information corresponding to the focus point and the historical information are consistent; the inference result includes semantic consistency and semantic inconsistency.
4. An incremental content recommendation device, characterized in that: include: A focus collection unit, used to collect user's focus; wherein the user's focus is described in natural language; A latest content acquisition unit, used to acquire the original content of the latest information of the channel from a preset channel; An information summarization unit, configured to apply a preset first reasoning model to summarize the original content according to the focus point to obtain summary information corresponding to the focus point; The first duplicate checking unit is used to determine the query condition according to the focus point when the summary information corresponding to the focus point is not empty, and query the historical information that meets the query condition in a preset database according to the query condition; the database is used to store historical information from various channels that matches the user's focus point; if no historical information that meets the query condition is found in the database, the database update and information push unit is triggered; or, if historical information that meets the query condition is found in the database, the second duplicate checking unit is triggered; A second duplication checking unit is used to apply a preset second reasoning model to infer whether the semantics expressed by the summary information corresponding to the focus point and the historical information are consistent; if the reasoning result is that the semantics expressed by the summary information corresponding to the focus point and the historical information are inconsistent, triggering a database update and information push unit; A database updating and information pushing unit, configured to update the summary information corresponding to the focus point to the database according to the query condition, and send the original content and the summary information to the user; Wherein, the information summarization unit includes: A focus prompt generating module, used for generating a prompt sentence corresponding to the focus point according to the focus point of the user; the prompt sentence corresponding to the focus point is used to instruct the first reasoning model to try to summarize the content matching the focus point from the original content in a natural language description manner; A first inference content aggregation module, used to aggregate the original content and the prompt sentence corresponding to the focus point to obtain a first content to be inferred; wherein the original content includes text, video, picture and / or audio; The reasoning and summarization module is used to input the first content to be inferred into the first reasoning model for reasoning. When there is content matching the focus point in the first content to be inferred, the first reasoning model summarizes the content matching the focus point from the first content to be inferred in a natural language description as the summary information corresponding to the focus point; when there is no content matching the focus point in the first content to be inferred, the summary information corresponding to the focus point is empty.
5. The incremental content recommendation device according to claim 4, characterized in that: The first duplicate checking unit is also used for: When the summary information corresponding to the focus point is empty, the original content is discarded and is not sent to the user.
6. The incremental content recommendation device according to claim 4, characterized in that: The second duplicate checking unit comprises: A second reasoning content generation module is used to merge the summary information corresponding to the focus point, the historical information and a preset prompt sentence into a second content to be reasoned; the preset prompt sentence is used to instruct the second reasoning model to reason whether the semantics of the summary information corresponding to the focus point and the historical information are consistent in a natural language description; The duplicate checking reasoning module is used to input the second content to be inferred into the second reasoning model for reasoning, and obtain the reasoning result of whether the semantics of the summary information corresponding to the focus point and the historical information are consistent; the reasoning result includes semantic consistency and semantic inconsistency.
7. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the incremental content recommendation method according to any one of claims 1 to 3 is implemented.
8. An electronic device, characterized in that: It includes: processor; a memory for storing instructions executable by the processor; The processor is configured to execute the instructions to implement the incremental content recommendation method as described in any one of claims 1-3.
