CN118708808A - Recommendation method, device, equipment and storage medium based on large model - Google Patents

Abstract

The present disclosure proposes a recommendation method, device, equipment and storage medium based on a big model, which relates to the field of data processing technology, especially to the field of natural language processing and big data technology. The specific implementation scheme is: by obtaining the historical behavior information of the target user; according to the historical behavior information, using the big model to determine the user portrait of the target user; based on the user portrait and the real-time behavior information of the target user, using the big model to determine the target task to be executed; recommending the client logged in by the target user to execute the target task. By generating a user portrait to extract the representative information of the target user, and then combining it with the real-time behavior information of the target user, the behavior information of the target user can be described more accurately, which can be used as the basis for determining the target task, which can improve the relevance and accuracy of the target task recommendation, personalize the work content, and improve the speed of the target user to perform the task.

Description

Recommendation method, device, equipment and storage medium based on large model

Technical Field

The present disclosure relates to the field of data processing technology, and in particular to the field of natural language processing and big data technology.

Background Art

In corporate office scenarios, employees often need to find and obtain a lot of information and make various decisions when handling daily work tasks. Most existing office automation systems rely on passive user operations, that is, users need to actively search or query relevant information, and the recommendation process is less intelligent.

Summary of the invention

The present disclosure provides a recommendation method, apparatus, device and storage medium based on a large model.

According to a first aspect of the present disclosure, a recommendation method based on a big model is provided, the method comprising: obtaining historical behavior information of a target user; determining a user portrait of the target user using a big model based on the historical behavior information; determining a target task to be executed using the big model based on the user portrait and the real-time behavior information of the target user; and recommending the client logged in by the target user to execute the target task.

According to a second aspect of the present disclosure, a recommendation device based on a big model is provided, comprising: an acquisition module for acquiring historical behavior information of a target user; a first determination module for determining a user portrait of the target user using a big model based on the historical behavior information; a second determination module for determining a target task to be executed using the big model based on the user portrait and real-time behavior information of the target user; and a recommendation module for recommending that a client logged in by the target user execute the target task.

According to a third 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 method described in the first aspect.

According to a fourth aspect of the present disclosure, a non-transitory computer-readable storage medium storing computer instructions is provided, wherein the computer instructions are used to cause the computer to execute the method described in the first aspect.

According to a fifth aspect of the present disclosure, a computer program product is provided, comprising a computer program, wherein the computer program implements the steps of the method according to the first aspect when executed by a processor.

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

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

FIG1 is a schematic diagram of a flow chart of a recommendation method based on a large model according to an embodiment of the present disclosure;

FIG2 is a flow chart of a large model-based recommendation method according to another embodiment of the present disclosure;

FIG3 is a flow chart of a large model-based recommendation method according to another embodiment of the present disclosure;

FIG4 is a flow chart of a large model-based recommendation method according to another embodiment of the present disclosure;

FIG5 is a schematic diagram of a structure of a recommendation device based on a large model according to an embodiment of the present disclosure;

FIG. 6 is a schematic block diagram of an example electronic device 600 that may be used to implement embodiments of the present disclosure.

DETAILED DESCRIPTION

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.

Data processing is to extract and derive valuable and meaningful data for certain people from a large amount of chaotic and difficult to understand data. Data is a form of expression of facts, concepts or instructions, which can be processed by manual or automatic devices. After the data is interpreted and given a certain meaning, it becomes information. Data processing is the collection, storage, retrieval, processing, transformation and transmission of data.

Natural Language Processing (NLP) is an important direction in the fields of computer science and artificial intelligence. It is a variety of theories and methods that can realize effective communication between people and computers using natural language. It is mainly used in machine translation, public opinion monitoring, automatic summarization, opinion extraction, text classification, question answering, text semantic comparison, speech recognition, Chinese OCR and other aspects.

Big Data refers to the amount of data involved that is so large that it cannot be captured, managed, processed, and organized into information that helps business decisions more proactively within a reasonable time through mainstream software tools, including structured, semi-structured, and unstructured data.

It should be noted that in the technical solution of the present disclosure, the collection/collection, update, analysis, use, transmission, storage and other aspects of the historical behavior information, user portraits and real-time behavior information involved are all permitted and agreed by the user, and comply with the provisions of relevant laws and regulations, and are used for legal and reasonable purposes. They are not shared, leaked or sold outside of these legal uses, and are subject to supervision and management by national regulatory authorities. Necessary measures should be taken for the above information to selectively block the use or access to the above information data to prevent illegal access to such information data, ensure that persons who have access to the above information data comply with the provisions of relevant laws and regulations, and ensure user information security. In addition, once the above information data is no longer needed, the risk should be minimized by limiting or even prohibiting data collection and/or deleting data.

FIG1 is a flow chart of a large model-based recommendation method provided by an embodiment of the present disclosure. As shown in FIG1 , the large model-based recommendation method includes but is not limited to the following steps:

Step 101, obtaining historical behavior information of the target user.

Optionally, the target user may be an enterprise employee, and the target user performs daily work by logging into a client on a terminal device.

In the embodiment of the present disclosure, after the target user confirms that historical behavior information is allowed to be obtained, the historical behavior information of the target user is obtained by retrieving the data of the client through which the target user logs in, wherein the historical behavior information is information corresponding to the target user's historical work behavior, such as the time, place, object, event, and other information corresponding to the target user's meeting behavior, project tracking behavior, customer communication behavior, email exchange behavior, document operation behavior, and other historical work behaviors.

For example, the retrieved target user data is: Employee A writes a weekly report on Friday. Then the target user is Employee A, the historical work behavior is writing a weekly report, the corresponding time is Friday, the location is work location B where Employee A writes the weekly report, and the object is Employee A. The historical behavior information corresponding to the data includes: Friday, work location B, Employee A, and writing a weekly report.

For example, the retrieved target user data is: On Thursday, employee A held a meeting with employees C, D, and E in his group. Then the target user is employee A, the historical work behavior is holding a meeting, the corresponding time is Thursday, the location is the work location F where the meeting was held, the objects are employees A, C, D, and E, and the historical behavior information corresponding to the data includes: Thursday, work location F, employee A, employee C, employee D, employee E, and holding a meeting.

Step 102, based on historical behavior information, a large model is used to determine the user profile of the target user.

In the disclosed embodiment, after the target user confirms permission to obtain relevant information of the user portrait, the acquired historical behavior information can be input into the big model, and the historical behavior information of the target user can be analyzed through the model capabilities of the big model to generate a user portrait of the target user.

Optionally, the user portrait can be representative information extracted by abstracting the historical behavior information of the target user within a certain period of time, for example, it can be an abstract extraction of all historical behavior information of employee A since he joined the company, or an abstract extraction of historical behavior information of employee A in the past six months.

It should be noted that the large model proposed in the embodiments of the present disclosure may be a large language model or other large models that can realize the determination of user portraits, and no specific limitation is made here.

Among them, the Large Language Model (LLM) is a deep learning model trained based on massive text data. It can generate natural language text, deeply understand the meaning of the text, and handle various natural language tasks such as text summarization, question answering, and translation.

Step 103, based on the user portrait and the real-time behavior information of the target user, a large model is used to determine the target task to be executed.

In the embodiment of the present disclosure, after the target user confirms permission to obtain the real-time behavior information, the real-time behavior information of the target user can be obtained by retrieving data of the client that the target user logs in to.

Optionally, the user portrait is a representation of the historical information of the target user, and the real-time behavior information is a representation of the current information of the target user. By combining the user portrait and the real-time behavior information, the behavior information of the target user can be comprehensively monitored.

In the disclosed embodiment, a large model is used to determine the target task to be performed by the target user from a variety of candidate tasks executable by the client logged in by the target user based on the user portrait and real-time behavior information of the target user.

Step 104: recommend the client logged in by the target user to perform the target task.

Optionally, the target task can be pushed to the client logged in by the target user in the form of a pop-up window, email, etc., to remind the target user to perform the target task.

The recommendation method based on a big model proposed in the present disclosure obtains the historical behavior information of the target user; determines the user portrait of the target user based on the historical behavior information; determines the target task to be executed based on the user portrait and the real-time behavior information of the target user using the big model; and recommends the client logged in by the target user to execute the target task. By generating a user portrait to extract representative information of the target user, and then combining the real-time behavior information of the target user, the historical information and the current information are combined, so that the behavior information of the target user can be described more accurately, and this can be used as the basis for determining the target task, which can improve the relevance and accuracy of the target task recommendation, improve the speed of the target user to execute the task, reduce the complexity of the target user's task planning operation, personalize the work content, and improve the target user's operation experience for office tools.

The present disclosure also proposes another recommendation method based on a large model. FIG2 is a flow chart of another recommendation method based on a large model provided by an embodiment of the present disclosure. As shown in FIG2, the recommendation method based on a large model includes but is not limited to the following steps:

Step 201, obtaining historical behavior information of the target user.

For the description of the above step 201, please refer to step 101 in the above embodiment, which will not be repeated here.

Step 202: Acquire a feature library constructed based on the preset behavior information, where the feature library is used to store the behavior elements of the preset behavior information and the associations between the behavior elements.

Optionally, the feature library can be constructed by the following steps:

Extract at least one behavior element of preset behavior information, where the behavior element includes at least one of the following: execution time, execution location, execution object and behavior type of the preset behavior; for any preset behavior information, determine the association relationship between the behavior elements; store the behavior element of any preset behavior information and the corresponding association relationship as a record to obtain a feature library.

Optionally, the preset behavior may be a pre-set behavior under multiple regular scenarios, for example, all employees in department 1 hold a weekly meeting every Friday, all employees in department 2 jointly edit work record documents every Thursday, and all employees arrive at the company before 10 am every Monday.

In the disclosed embodiment, preset behaviors in various scenarios are preset according to enterprise operation regulations, and a feature library is constructed by extracting the behavior elements of preset behavior information corresponding to each preset behavior and the correlation between the behavior elements.

For example, if the preset behavior is: all employees in department 1 hold a weekly meeting every Friday, the extracted behavior elements include: every Friday, weekly meeting location, all employees in department 1, and weekly meeting. A serial number is generated for the preset behavior, and each behavior element is used as the data corresponding to the serial number. The association relationship of each behavior element is generated to form a record, which is converted into natural language type data for storage. The records corresponding to each preset behavior form a feature library.

By building a feature library based on preset behavioral information, the company's regular scenarios can be counted and monitored, providing a data basis for generating user portraits for target users, so as to summarize and coordinate the behavioral information of target users.

Step 203, matching the historical behavior information in the feature library through the big model to obtain the behavior elements and associations corresponding to the historical behavior information; the behavior elements and associations corresponding to the historical behavior information are used as the user profile of the target user.

In the disclosed embodiment, the historical behavior information of the target user includes the time, place, object, event and other information corresponding to the historical work behavior. The historical behavior information is queried and matched in the feature library through the big model, and the matching result is used to determine whether the historical work behavior belongs to the preset behavior. When the historical work behavior belongs to the preset behavior, the behavior elements and associations corresponding to the historical behavior information are determined, that is, the records corresponding to the historical behavior information in the feature library are determined.

Optionally, the behavior elements and associations corresponding to the historical behavior information of the target user are used as the user profile of the target user.

For example, employee A is the target user and belongs to Department 1. His historical behavior information includes weekly meetings on multiple Fridays. The corresponding behavioral elements matched in the feature library are every Friday, weekly meeting location, all employees in Department 1, and weekly meetings. This record is written into the user portrait of employee A as data corresponding to a serial number.

Optionally, the behavior elements and associations obtained after matching the historical behavior information are sequentially written into the user portrait of the target user, and finally the user portrait of the target user is obtained.

It should be noted that in the embodiments of the present disclosure, matching can be performed through Embedding, or other methods that can achieve text matching, which are not specifically limited here.

The user portrait of the target user is generated through feature library matching, including information such as the target user's behavioral habits, work content preferences, and precautions. It can abstract the representative information of the target user from multiple dimensions, gain a comprehensive understanding of the target user, and accurately locate the target user based on historical work content.

Optionally, the user portrait of the target user may also include an account identifier generated for the target user, such as a title such as engineer or technical expert, which is used as a title when interacting with the target user.

In the disclosed embodiment, the generated user portrait is stored in TimelyProfile, and the generated title is stored in UserInputProfile, wherein TimelyProfile and UserInputProfile are different storage fields in the same storage database.

Among them, user portraits are stored in natural language form and lightweight data exchange JSON format.

By extracting multi-dimensional information of the target user through historical behavior information, we can abstract the representative information of the target user. The generated user portrait can accurately describe the historical work information of the target user, gain a comprehensive understanding of the target user, and accurately locate the target user based on the historical work content.

Step 204, based on the user portrait and the real-time behavior information of the target user, a large model is used to determine the target task to be executed.

Step 205: recommend the client logged in by the target user to perform the target task.

For the description of the above step 204 and step 205, please refer to step 103 and step 104 in the above embodiment, which will not be repeated here.

The recommendation method based on a big model proposed in the present disclosure obtains the historical behavior information of the target user; obtains a feature library constructed based on the preset behavior information, the feature library is used to store the behavior elements of the preset behavior information and the associations between the behavior elements; matches the historical behavior information in the feature library through the big model to obtain the behavior elements and associations corresponding to the historical behavior information; uses the behavior elements and associations corresponding to the historical behavior information as the user portrait of the target user; based on the user portrait and the real-time behavior information of the target user, uses the big model to determine the target task to be performed; and recommends the client logged in by the target user to perform the target task. Through the historical behavior information, multi-dimensional information extraction of the target user is performed to generate a user portrait that can accurately describe the historical work information of the target user. Combined with the real-time behavior information of the target user, the behavior information of the target user can be adjusted in real time. This is used as the basis for determining the target task, which can improve the relevance and accuracy of the target task recommendation and improve the target user's operating experience for office tools.

The present disclosure also proposes another recommendation method based on a large model. FIG3 is a flow chart of another recommendation method based on a large model provided by an embodiment of the present disclosure. As shown in FIG3, the recommendation method based on a large model includes but is not limited to the following steps:

Step 301, obtaining historical behavior information of the target user.

Step 302: Based on the historical behavior information, a large model is used to determine the user profile of the target user.

For the description of the above step 301 and step 302, please refer to step 101 and step 102 in the above embodiment, which will not be repeated here.

Step 303: Generate a first prompt word based on the user portrait and real-time behavior information; based on the first prompt word, use the big model to call the user portrait update tool to update the user portrait.

Optionally, after the big model monitors the latest real-time behavior of the target user, a first prompt word is generated based on the user portrait and real-time behavior information. The first prompt word is used to instruct the big model to initiate thinking to determine whether the real-time behavior will cause changes to the user portrait. When it is determined that the real-time behavior will cause changes to the user, the user portrait update tool is called to update the user portrait.

The first prompt word includes at least role information, instructions, information about available tools, restrictions on the use of all tools, historical behavior information, output format of the large model output results, and the current task.

As an example, after the target user has the latest real-time behavior, the first prompt word generated is:

Character Information:

Your name is: User Profile Update Assistant

Your task description is: Based on the user's behavior and the user's current user profile, determine whether the user's user profile needs to be updated.

Your instruction is:

You have a few things to think about:

1. Does this behavior reveal new interests or preferences?

2. Does this behavior show a change in user behavior patterns?

3. Does this behavior require updating the personal attributes in the portrait?

If it is determined that an update is necessary, use the user portrait update tool to update the portrait.

Information about available tools:

The only tools you can use are:

[User portrait update tool]: Update the user's user portrait information based on the user's current user portrait and current behavior.

All tool usage restrictions:

1. Please try to reduce the number of tool calls while meeting user requirements.

2. Only select candidate tools and only use the information provided to you!

Output format requirements:

1. When the tool must be used, output in the following format:

Thought: [Reason for choosing this tool, no more than 30 words]

Action: [Tool name. Do not make it up]

Action Input:[Tool parameters]

Observation:[Tool execution results]

2. When the task is completed or no tool needs to be called, the output is in the following format:

Thought:[output "task completed"]

Final Answer: [Please output the answer to the user in markdown format. It must follow the facts and provide users with complete solutions and suggestions.]

start!

Historical behavior information: [Multiple rounds of historical behavior information are sorted from new to old for reference.]

[history]

Current tasks:

User information: [name]

Current time: [time]

Real-time behavior information: [user_input]

Update user profile: [user_profile]

Thought:

In this example, role information is used to instruct the big model to confirm role information, instructions are used to instruct the big model to execute instructions, available tool information is used to instruct the big model to select tool calls, all tool usage restrictions are used to restrict the conditions for the big model to call tools, historical behavior information is used as a reference condition for the output results of the big model, the current task includes user portraits and real-time behavior information, and the output format of the big model output results is used to specify the output format corresponding to different judgment results.

In the disclosed embodiment, the first prompt word is input into the big model so that the big model determines whether the real-time behavior will change the user portrait, and calls the user portrait update tool when it is determined that the real-time behavior will change the user.

As an example, calling the user profile update tool can be achieved through the prompt word:

The prompt words of the user portrait update tool are:

Your task is to update the user profile based on the user's behavior, the user's current profile, and previous thinking information.

User Action: [user_input]

Current user profile: [user_profile]

Thought information: [thought]

Notice:

1. This task has already determined whether a portrait update is required, so please update the portrait based on the above information and do not refuse or perform other operations.

2. User portraits cannot contain sensitive information such as pornography, gambling, drugs, violence, and politics.

3. User portraits should remain objective and should not contain subjective evaluation information about users.

4. Do not summarize profile information that is worthless or meaningless.

Based on the above information and requirements, please directly output the updated user portrait.

Updated user profile:

Among them, the content of attention can be understood as the reference conditions of the output results of the large model.

In the disclosed embodiment, the user portrait update tool updates the user portrait based on the thinking results of the user portrait and the big model, and makes real-time adjustments to the behavior information of the target user to obtain the most timely and accurate description of the target user.

As an example, updating a user profile can be achieved through the following steps:

When there are behavioral elements and associations matching the real-time behavioral information in the feature library, and there are no behavioral elements and associations matching the real-time behavioral information in the user portrait, the behavioral elements and associations corresponding to the real-time behavioral information are written into the user portrait.

By writing behavioral elements and associations of real-time behavioral information that do not exist in the user portrait, the user portrait can be updated in real time. This allows personalized adjustments to the user portrait of the target user, timely control of changes in the target user's related work, and acquisition of dynamic user portraits to provide accurate data support for the determination of target tasks.

Furthermore, the big model also has the ability to reflect, and prompt words can be used to instruct the big model to reflect on its own output results.

As an example, the reflection prompts are:

You are a reflection assistant. Your current task is to determine whether the update behavior is successful and effective based on the user's current profile, the user's updated profile, and the user's behavior.

User Action: [user_input]

Current user profile: [user_profile]

User updated profile: [updated_user_profile]

Please output your judgment and thinking in the format of thinking first and then judging. Please output whether the update is successful or failed.

think:

judge:

Through reflection prompts, the large model can think and judge the output results to determine the validity of the output results.

As another example, when a tool called by the large model reports an error, the operation history and operation error information are returned to the large model so that the large model can reflect and correct the error based on the error information.

Step 304 , based on the updated user profile, the target task is screened in the recommendation candidate set through the large model, wherein the recommendation candidate set stores a variety of candidate tasks executable by the client logged in by the target user.

Optionally, a second prompt word is generated based on the updated user portrait; based on the second prompt word, the large model is used to call the target task screening tool and the sorting tool to determine the target task to be executed.

After determining the second prompt word, the large model can call the tool through the prompt word to determine the target task to be executed.

In the disclosed embodiment, a large model is used to call a target task screening tool to extract semantic information of the updated user portrait based on the second prompt word and map it to a first vector; the semantic information of any candidate task in the recommended candidate set is extracted and mapped to a second vector; the large model is used to call a sorting tool to sort the similarity between the first vector and the second vector; and a target number of target tasks are selected according to the similarity sorting.

Optionally, the second prompt word includes at least role information, instructions, information about available tools, restrictions on the use of all tools, output format requirements, and the current task.

As an example, the second prompt word may be:

Character Information:

Your name is: Active Recommendation Assistant

Your task description is: Based on the user's behavior, use several tools reasonably to ultimately recommend the results that the user is most likely to be interested in or related to.

Your instruction is:

Based on the user's current behavior, determine which tools should be used in sequence and finally recommend results to the user. It is also possible that no recommendation is needed and you can end the task directly.

Please return the information strictly according to the requirements.

Information about available tools:

The only tools you can use are:

[User portrait filtering tool]: Get the updated user portrait of the target user. The user portrait filtering tool will also filter out the portrait information that is more relevant to the current behavior from the user portrait information.

[Target Task Screening Tool]: Based on known information, filter out the content that should be recommended to the user from the recommendation candidate set.

[Sorting tool]: Sort the candidate recommendation results according to user portrait information.

All tool usage restrictions:

1. Please try to reduce the number of tool calls while meeting user requirements.

2. Only select candidate tools and only use the information provided to you!

Output format requirements:

1. When the tool must be used, output in the following format:

Thought: [Reason for choosing this tool, no more than 30 words]

Action: [Tool name. Do not make it up]

Action Input:[Tool parameters]

Observation:[Tool execution results]

2. When the task is completed or no tool needs to be called, the output is in the following format:

Thought:[output "task completed"]

Final Answer: [Please output the answer to the user in markdown format. It must follow the facts and provide users with complete solutions and suggestions.]

start!

Historical user behavior information: [Multiple rounds of historical behaviors are sorted from new to old for reference.]

[history]

Current tasks:

User information: [name]

Current time: [time]

Real-time behavior information: [user_input]

Current user profile: [user_profile]

Thought:

In this example, role information is used to instruct the big model to confirm role information, instructions are used to instruct the big model to execute instructions, available tool information is used to instruct the big model to select tool calls, all tool usage restrictions are used to restrict the conditions for the big model to call tools, historical behavior information is used as a reference condition for the output results of the big model, the current task includes the updated user portrait (user's current portrait) and real-time behavior information, and the output format of the big model's output results is used to specify the output format corresponding to different judgment results.

Among them, the target task screening tool is used to: extract the semantic information of the updated user portrait and map it into a first vector; extract the semantic information of any candidate task in the recommended candidate set and map it into a second vector; determine the target task based on the similarity between the first vector and any second vector.

As an example, the prompt word for calling the target task screening tool is:

You are a recommendation content screening tool. Please filter out the content that should be recommended to users based on user portraits and user behaviors.

User information: [name]

Current time: [time]

User's current action: [user_input]

Current user profile: [user_profile]

Recommended candidate set: [data]

Based on the above information, select [n] results from the recommendation candidate set. Please output them in the original format as the content that should be recommended to users, and output your thoughts first.

think:

Filtered recommendations:

The recommendation candidate set includes a variety of recommended tasks that can be executed by the client logged in by the target user.

By calling the target task screening tool, the target task can be determined based on the similarity between the semantic information of the updated user portrait and the semantic information of any candidate task. The most suitable target task for the target user can be screened out through the correlation between the recommended candidate set and the updated user portrait, and task recommendations can be made in a timely manner, reducing the target user's search and waiting time, and improving the target user's operating experience for office tools.

The sorting tool is used to: sort the similarities; select the target number of target tasks according to the similarity sorting.

As an example, the similarities are sorted in descending order, candidate tasks corresponding to the TOP-N similarities are selected as target tasks, and the target tasks are recommended and arranged in the order of similarity.

As another example, a similarity threshold is preset, and candidate tasks corresponding to similarities greater than the similarity threshold are selected as target tasks. When there are multiple target tasks, a sorting tool is called to sort the similarities corresponding to the target tasks according to numerical values, and target tasks with higher similarities are recommended as more relevant target tasks in a priority manner. This can improve the relevance and accuracy of target task recommendations, or help target users arrange their work content reasonably, thereby reducing the operational complexity of task planning for target users.

By updating the user portrait in real time through real-time behavior information, the user portrait of the target user can be personalized to obtain a dynamic, detailed and accurate user portrait, which can be used as data support for determining the target task and improve the relevance and accuracy of the target task recommendation.

By calling multiple tools to screen target tasks, it is possible to flexibly select corresponding tools for task processing according to different scenarios, quickly execute related tasks in complex scenarios, improve execution capabilities, select one or more target tasks that are most suitable for the target user from the recommended candidate set, and intelligently provide relevant target tasks for the target user.

Furthermore, before generating the second prompt word, the updated user portrait may be preliminarily screened.

As an example, the prompt words for the user portrait screening tool are:

You are a user portrait screening tool that screens out the more relevant user portraits based on user behavior.

User information: [name]

Current time: [time]

Real-time behavior information: [user_input]

User profile: [user_profile]

Require:

Filter out portraits that are relatively relevant or potentially relevant to the behavior, and filter out completely irrelevant portraits.

On this basis, the richness and completeness of the portrait information is guaranteed.

Please output the filtered image as required:

User portrait after screening:

By screening the updated user portraits and selecting more relevant portrait content, we can further refine the work status of the target users.

Furthermore, after obtaining the user portrait, the initial task can be selected as the target task from the recommended candidate set through the large model based on the user portrait corresponding to the historical behavior data. If there are multiple initial tasks, they can be sorted through Embedding.

By selecting the initial tasks, cold start recommendation content is generated for target users while focusing on speed, stability and cost, thus ensuring the initial experience of target users.

Step 305: recommend the client logged in by the target user to perform the target task.

For the description of the above step 305, please refer to step 104 in the above embodiment, which will not be repeated here.

The recommendation method based on the big model proposed in the present disclosure obtains the historical behavior information of the target user; determines the user portrait of the target user using the big model according to the historical behavior information; generates the first prompt word based on the user portrait and real-time behavior information; calls the user portrait update tool based on the first prompt word using the big model to update the user portrait; screens the target task from the recommendation candidate set based on the updated user portrait through the big model; and recommends the client logged in by the target user to perform the target task. By updating the user portrait in real time, the user portrait of the target user can be personalized, the changes in the relevant work of the target user can be controlled in time, and a dynamic, detailed and accurate user portrait can be obtained as the basis for determining the target task, which can improve the relevance and accuracy of the target task recommendation and improve the target user's operating experience for office tools.

The present disclosure also proposes another recommendation method based on a large model. FIG4 is a flow chart of another recommendation method based on a large model provided by an embodiment of the present disclosure. As shown in FIG4, the recommendation method based on a large model includes but is not limited to the following steps:

Step 401, obtaining historical behavior information of the target user.

Step 402: Acquire a feature library constructed based on the preset behavior information, where the feature library is used to store the behavior elements of the preset behavior information and the associations between the behavior elements.

Step 403, the historical behavior information is matched in the feature library through the big model to obtain the behavior elements and associations corresponding to the historical behavior information; the behavior elements and associations corresponding to the historical behavior information are used as the user profile of the target user.

Step 404, based on the user portrait and real-time behavior information, a first prompt word is generated; based on the first prompt word, a user portrait update tool is called using a large model to update the user portrait.

Step 405 , based on the updated user portrait, the target task is screened in the recommendation candidate set through the large model, wherein the recommendation candidate set stores a variety of candidate tasks executable by the client logged in by the target user.

Step 406: recommend the client logged in by the target user to perform the target task.

For the description of the above steps 401 to 406, please refer to the relevant description in the above embodiment, which will not be repeated here.

It should be noted that in an enterprise office scenario, the work tasks handled by the target users include but are not limited to file management, meeting scheduling, project progress tracking, knowledge acquisition, and communication and collaboration.

The recommendation method based on the big model proposed in this disclosure can actively recommend relevant files and materials according to the work content and habits of the target user, simplify the search process; intelligently remind the appropriate meeting time and relevant participants based on the target user's schedule and work focus to optimize the meeting arrangement; timely remind the target user to pay attention to important project progress and milestones to ensure that the project is advanced on time; actively push relevant research results, industry trends and training materials according to the target user's professional field and current tasks to improve the target user's knowledge level; analyze the target user's communication records and cooperative relationships, recommend potential partners and communication opportunities, and promote teamwork. Make corporate office more intelligent and promote the efficient operation of enterprises.

The big model-based recommendation method proposed in the present disclosure obtains the historical behavior information of the target user; obtains a feature library constructed based on the preset behavior information, the feature library is used to store the behavior elements of the preset behavior information and the associations between the behavior elements; matches the historical behavior information in the feature library through the big model to obtain the behavior elements and associations corresponding to the historical behavior information; uses the behavior elements and associations corresponding to the historical behavior information as the user portrait of the target user; generates a first prompt word based on the user portrait and real-time behavior information; based on the first prompt word, uses the big model to call the user portrait update tool to update the user portrait; based on the updated user portrait, uses the big model to filter the target task in the recommendation candidate set, wherein the recommendation candidate set stores a variety of candidate tasks executable by the client logged in by the target user; and recommends the client logged in by the target user to execute the target task. Through multi-dimensional information extraction of target users based on historical behavior information, a user portrait is generated that can accurately describe the target user's historical work information. The user portrait is further updated in real time, and personalized adjustments can be made to the user portrait of the target user. The target user's task operation changes can be controlled in a timely manner, and dynamic user portraits can be obtained. This is used as the basis for determining the target task, which can improve the relevance and accuracy of the target task recommendation and enhance the target user's operating experience for office tools.

An embodiment of the present disclosure further proposes a recommendation device based on a large model. Since the recommendation device based on a large model proposed in the embodiment of the present disclosure corresponds to the recommendation methods based on a large model proposed in the above-mentioned embodiments, the implementation methods of the above-mentioned recommendation methods based on a large model are also applicable to the recommendation device based on a large model proposed in the embodiment of the present disclosure, and will not be described in detail in the following embodiments.

Fig. 5 is a schematic diagram of a large model-based recommendation device according to an embodiment of the present disclosure. As shown in Fig. 5 , the large model-based recommendation device 500 includes: an acquisition module 501 , a first determination module 502 , a second determination module 503 and a recommendation module 504 .

Acquisition module 501, used to acquire historical behavior information of target users;

A first determination module 502 is used to determine a user profile of a target user using a large model based on historical behavior information;

The second determination module 503 is used to determine the target task to be executed using a large model based on the user portrait and the real-time behavior information of the target user;

The recommendation module 504 is used to recommend the client logged in by the target user to perform the target task.

In the embodiment of the present disclosure, the first determining module is used to:

Acquire a feature library constructed based on the preset behavior information, the feature library being used to store the behavior elements of the preset behavior information and the association relationship between the behavior elements;

The historical behavior information is matched in the feature library through the big model to obtain the behavior elements and associations corresponding to the historical behavior information;

The behavioral elements and associations corresponding to the historical behavioral information are used as the user profile of the target user.

In the embodiment of the present disclosure, the second determining module is used to:

Generate the first prompt word based on the user portrait and real-time behavior information;

Based on the first prompt word, the user portrait update tool is called by the large model to update the user portrait;

Based on the updated user profile, the target task is filtered out from the recommendation candidate set through the large model, where the recommendation candidate set stores a variety of candidate tasks that can be executed by the client.

When there are behavioral elements and associations matching the real-time behavioral information in the feature library, and there are no behavioral elements and associations matching the real-time behavioral information in the user portrait, the behavioral elements and associations corresponding to the real-time behavioral information are written into the user portrait.

Generate a second prompt word based on the updated user portrait;

Based on the second prompt word, the large model is used to call the target task screening tool and the sorting tool to determine the target task to be executed.

The target task screening tool is called by the large model to extract the semantic information of the updated user portrait based on the second prompt word, and is mapped into the first vector;

Extracting semantic information of any candidate task in the recommendation candidate set and mapping it into a second vector;

Using the large model to call a sorting tool to sort the similarity between the first vector and the second vector;

Select the target tasks with the target number sorted by similarity.

The recommendation device based on a big model proposed in the present disclosure obtains the historical behavior information of the target user; determines the user portrait of the target user based on the historical behavior information; determines the target task to be executed based on the user portrait and the real-time behavior information of the target user using the big model; and recommends the client logged in by the target user to execute the target task. By generating a user portrait to extract representative information of the target user, and then combining the real-time behavior information of the target user, the historical information and the current information, the behavior information of the target user can be described more accurately, which can be used as the basis for determining the target task, and can improve the relevance and accuracy of the target task recommendation, improve the speed of the target user to execute the task, reduce the complexity of the target user's task planning operation, personalize the work content, and improve the target user's operation experience for office tools.

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.

FIG6 shows a schematic block diagram of an example electronic device 600 that can be used to implement an embodiment of the present disclosure. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workbenches, 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.

As shown in FIG6 , the device 600 includes a computing unit 601, which can perform various appropriate actions and processes according to a computer program stored in a ROM (Read-Only Memory) 602 or a computer program loaded from a storage unit 608 into a RAM (Random Access Memory) 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 I/O (Input/Output) interface 605 is also connected to the bus 604.

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.

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 CPU (Central Processing Unit), a GPU (Graphic Processing Units), various dedicated AI (Artificial Intelligence) computing chips, various computing units running machine learning model algorithms, a DSP (Digital Signal Processor), and any appropriate processor, controller, microcontroller, etc. The computing unit 601 performs the various methods and processes described above, such as a recommendation method based on a large model. For example, in some embodiments, the recommendation method based on a large 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 recommendation method based on the large model described above may be performed. Alternatively, in other embodiments, the computing unit 601 may be configured to execute the large model-based recommendation method in any other appropriate manner (eg, by means of firmware).

Various embodiments of the systems and techniques described above herein may be implemented in digital electronic circuit systems, integrated circuit systems, FPGAs (Field Programmable Gate Arrays), ASICs (Application-Specific Integrated Circuits), ASSPs (Application Specific Standard Products), SOCs (System On Chips), CPLDs (Complex Programmable Logic Devices), computer hardware, firmware, software, and/or combinations thereof. These various embodiments may include: being implemented in one or more computer programs that may be executed and/or interpreted on a programmable system including at least one programmable processor that may be a dedicated or general-purpose programmable processor that may receive data and instructions from a storage system, at least one input device, and at least one output device, and may transmit data and instructions to the storage system, the at least one input device, and the at least one output device.

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, implements the functions/operations specified in the flow chart and/or block diagram. 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.

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 device, or any suitable combination of the foregoing. More specific examples of machine-readable storage media may include electrical connections based on one or more lines, portable computer disks, hard disks, RAM, ROM, EPROM (Electrically Programmable Read-Only-Memory) or flash memory, optical fiber, CD-ROM (Compact Disc Read-Only Memory), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

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).

The systems and techniques described herein may be implemented in a computing system that includes backend 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 frontend 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 backend components, middleware components, or frontend 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: LAN (Local Area Network), WAN (Wide Area Network), the Internet, and blockchain networks.

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 between the client and the server is generated by computer programs running on the corresponding computers and having a client-server relationship with each other. The server may be a cloud server, also known as a cloud computing server or cloud host, which is a host product in the cloud computing service system to solve the defects of difficult management and weak business scalability in traditional physical hosts and VPS services ("Virtual Private Server", or "VPS" for short). The server may also be a server of a distributed system, or a server combined with a blockchain.

It should be noted that artificial intelligence is a discipline that studies how computers can simulate certain human thought processes and intelligent behaviors (such as learning, reasoning, thinking, planning, etc.), and includes both hardware-level and software-level technologies. Artificial intelligence hardware technologies generally include technologies such as sensors, dedicated artificial intelligence chips, cloud computing, distributed storage, and big data processing; artificial intelligence software technologies mainly include computer vision technology, speech recognition technology, natural language processing technology, as well as machine learning/deep learning, big data processing technology, knowledge graph technology, and other major directions.

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.

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 (17)

1. A large model-based recommendation method, the method comprising:

Acquiring historical behavior information of a target user;

Determining a user portrait of the target user by adopting a large model according to the historical behavior information;

Determining a target task to be executed by adopting the large model based on the user portrait and the real-time behavior information of the target user;

And recommending the client logged in by the target user to execute the target task.

2. The method of claim 1, wherein said determining a user representation of the target user using a large model based on the historical behavior information comprises:

Acquiring a feature library constructed based on preset behavior information, wherein the feature library is used for storing behavior elements of the preset behavior information and association relations among the behavior elements;

matching the historical behavior information in the feature library through the large model to obtain behavior elements and association relations corresponding to the historical behavior information;

and taking the behavior elements and the association relations corresponding to the historical behavior information as user portraits of the target user.

3. The method of claim 2, wherein the obtaining a feature library constructed based on preset behavior information comprises:

extracting at least one behavior element of the preset behavior information, wherein the behavior element comprises at least one of the following components: presetting execution time, execution place, execution object and behavior type of a behavior;

determining the association relation between all behavior elements for any preset behavior information;

and storing the behavior elements of any preset behavior information and the corresponding association relations as a record to obtain the feature library.

4. The method of claim 2, wherein the determining the target task to be performed using the large model based on the user representation and real-time behavioral information of the target user comprises:

generating a first prompt word based on the user portrait and the real-time behavior information;

Based on the first prompt word, calling a user portrait updating tool by adopting the large model to update the user portrait;

And screening the target tasks in a recommendation candidate set through the large model based on the updated user portraits, wherein the recommendation candidate set stores a plurality of candidate tasks executable by the client.

5. The method of claim 4, wherein the invoking a user representation update tool with the large model based on the first prompt word to update the user representation comprises:

And writing the behavior elements and the association relations corresponding to the real-time behavior information in the user portrait under the condition that the behavior elements and the association relations matched with the real-time behavior information exist in the feature library and the behavior elements and the association relations matched with the real-time behavior information do not exist in the user portrait.

6. The method of claim 4, wherein the determining the target task to be performed using the large model based on the user representation and real-time behavioral information of the target user comprises:

Generating a second prompt word based on the updated user portrait;

And based on the second prompt word, calling a target task screening tool and a sequencing tool by adopting the large model, and determining a target task to be executed.

7. The method of claim 6, wherein the determining the target task to be performed based on the second prompt word using the large model to invoke a target task screening tool and a ranking tool comprises:

Invoking the target task screening tool by adopting the large model, extracting semantic information of the updated user portrait based on the second prompt word, and mapping the semantic information into a first vector;

extracting semantic information of any candidate task in the recommended candidate set and mapping the semantic information into a second vector;

invoking the ranking tool to rank the similarity between the first vector and the second vector using the large model;

and selecting the target tasks with target quantity according to the sequence of the similarity.

8. A large model-based recommendation device, comprising:

the acquisition module is used for acquiring historical behavior information of the target user;

the first determining module is used for determining the user portrait of the target user by adopting a large model according to the historical behavior information;

The second determining module is used for determining a target task to be executed by adopting the large model based on the user portrait and the real-time behavior information of the target user;

And the recommending module is used for recommending the client logged in by the target user to execute the target task.

9. The apparatus of claim 8, wherein the first determining module is configured to:

Acquiring a feature library constructed based on preset behavior information, wherein the feature library is used for storing behavior elements of the preset behavior information and association relations among the behavior elements;

matching the historical behavior information in the feature library through the large model to obtain behavior elements and association relations corresponding to the historical behavior information;

and taking the behavior elements and the association relations corresponding to the historical behavior information as user portraits of the target user.

10. The apparatus of claim 9, wherein the first determining module is further configured to:

extracting at least one behavior element of the preset behavior information, wherein the behavior element comprises at least one of the following components: presetting execution time, execution place, execution object and behavior type of a behavior;

determining the association relation between all behavior elements for any preset behavior information;

and storing the behavior elements of any preset behavior information and the corresponding association relations as a record to obtain the feature library.

11. The apparatus of claim 9, wherein the second determining module is configured to:

generating a first prompt word based on the user portrait and the real-time behavior information;

Based on the first prompt word, calling a user portrait updating tool by adopting the large model to update the user portrait;

And screening the target tasks in a recommendation candidate set through the large model based on the updated user portraits, wherein the recommendation candidate set stores a plurality of candidate tasks executable by the client.

12. The apparatus of claim 11, wherein the second determination module is further configured to:

And writing the behavior elements and the association relations corresponding to the real-time behavior information in the user portrait under the condition that the behavior elements and the association relations matched with the real-time behavior information exist in the feature library and the behavior elements and the association relations matched with the real-time behavior information do not exist in the user portrait.

13. The apparatus of claim 11, wherein the second determination module is further configured to:

Generating a second prompt word based on the updated user portrait;

And based on the second prompt word, calling a target task screening tool and a sequencing tool by adopting the large model, and determining a target task to be executed.

14. The apparatus of claim 13, wherein the second determination module is further configured to:

Invoking the target task screening tool by adopting the large model, extracting semantic information of the updated user portrait based on the second prompt word, and mapping the semantic information into a first vector;

extracting semantic information of any candidate task in the recommended candidate set and mapping the semantic information into a second vector;

invoking the ranking tool to rank the similarity between the first vector and the second vector using the large model;

and selecting the target tasks with target quantity according to the sequence of the similarity.

15. An electronic device, comprising:

At least one processor; and

A memory communicatively coupled to the at least one processor; wherein,

The memory stores instructions executable by the at least one processor to enable the at least one processor to perform the method of any one of claims 1-7.

16. A non-transitory computer readable storage medium storing computer instructions for causing the computer to perform the method of any one of claims 1-7.

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