Recommendation method and device of undertaking unit and server

Abstract

The invention provides an undertaking unit recommendation method and device and a server, and relates to the technical field of big language model recommendation, and the method comprises the steps: obtaining work order information and a preset undertaking unit set, and the work order information comprises the work order content and the work order type; sending the work order content and the work order type to a multi-path recommendation model based on traditional language processing, performing coarse recall processing on each undertaking unit in a preset undertaking unit set, and determining a target undertaking unit set; and based on the business mixed thinking chain prompt project and the big language decision model, screening the target undertaking unit set, and determining a target undertaking unit which is the most recommended undertaking unit for solving the problem corresponding to the input work order. According to the invention, the recommendation accuracy of the undertaking unit can be obviously improved.

Description

Technical Field

[0001] The present invention relates to the technical field of large language model recommendation, and in particular to a recommendation method, device and server for an organizing unit. Background Art

[0002] The long-tail problem refers to the situation in which most of the data is concentrated in a few categories, while the majority of the remaining categories have only a small amount of data. This phenomenon of unbalanced data distribution is widely present in many practical applications. At present, relevant technologies have proposed that in the process of constructing work orders for citizen hotline content and setting accurate labels to determine the undertaking unit to solve the work order problem, the undertaking unit can be recommended through traditional classifier training and similarity calculation. However, the undertaking unit recommendation scheme lacks the combination of business and large models in the government field, cannot solve the long-tail problem, and the accuracy of the undertaking unit recommendation is low. Summary of the invention

[0003] In view of this, an object of the present invention is to provide a method, device and server for recommending an organizing unit, which can significantly improve the accuracy of the organizing unit's recommendation.

[0004] In the first aspect, an embodiment of the present invention provides a method for recommending a sponsoring unit, the method comprising: obtaining work order information and a preset sponsoring unit set, wherein the work order information comprises: work order content and work order type; sending the work order content and work order type to a multi-channel recommendation model based on traditional language processing, performing a rough recall process on each sponsoring unit in the preset sponsoring unit set, and determining a target sponsoring unit set; based on a business hybrid thinking chain prompting project and a large language decision model, screening the target sponsoring unit set and determining a target sponsoring unit, wherein the target sponsoring unit is the most recommended sponsoring unit for solving the problem corresponding to the input work order.

[0005] In one embodiment, the work order content and work order type are sent to a multi-channel recommendation model based on traditional language processing, and a rough recall process is performed on each undertaking unit in a preset undertaking unit set. The step of determining a target undertaking unit set includes: after determining the work order content and work order type, a rough recall process is performed on each undertaking unit through each recall strategy in the multi-channel recommendation model respectively, and the recommended undertaking unit corresponding to each recall strategy is determined, and the set of each recommended undertaking unit is determined as the target undertaking unit set, wherein the various recall strategies in the multi-channel recommendation model include: business rule recall strategy, classifier recall strategy, point problem classification recall strategy, event recall strategy, undertaking unit responsibility recall strategy and similar work order content recall strategy.

[0006] In one embodiment, each recall strategy in the multi-path recommendation model is used to perform a rough recall process on each undertaking unit, and the steps of determining the recommended undertaking unit corresponding to each recall strategy include: matching the category to which the work order content belongs through preset standardized business rules, and determining the recommended undertaking unit corresponding to the business rule recall strategy; matching the work order content with a classifier based on the historical number of recalls of the undertaking unit, and determining the recommended undertaking unit corresponding to the classifier recall strategy; vectorizing the work order content according to the work order type and the point information in the work order content, and determining the undertaking unit with the highest similarity with the point information of the work order in the preset undertaking unit set as the recommended undertaking unit corresponding to the point problem classification recall strategy through dot product similarity calculation.

[0007] In one embodiment, the method also includes: extracting events for historical work orders and constructing an event vector database to match the undertaking unit corresponding to the work order with the highest similarity in the event vector database, and determine it as the recommended undertaking unit corresponding to the event recall strategy; obtaining a portrait of the undertaking unit, and determining the recommended undertaking unit corresponding to the undertaking unit responsibility recall strategy by similarity matching the keywords in the work order content with the keywords in the undertaking unit portrait; vectorizing the work order content within a preset time threshold, and performing similarity recall through text content vectorization and motor calculation to determine the recommended undertaking unit corresponding to the similar work order content recall strategy.

[0008] In one embodiment, based on the business hybrid thinking chain prompt project and the big language decision model, the target hosting unit set is screened and processed, and the step of determining the target hosting unit includes: generating input prompts corresponding to the target hosting unit set through the business hybrid thinking chain prompt project; and determining the target hosting unit from the target hosting unit set based on the input prompts through the big language decision model.

[0009] In one embodiment, the step of generating input prompts corresponding to the target hosting unit set through a business hybrid thinking chain prompt project includes: after marking the source of the target hosting unit set, automatically generating input prompts corresponding to each hosting unit through the business hybrid thinking chain prompt project, wherein the input prompts include: background, role, task, main content of the work order and reason for recommending the hosting unit.

[0010] In one embodiment, the step of determining a target organizer from a set of target organizers based on an input prompt through a large language decision model includes: performing a confidence assessment on each organizer in the set of target organizers according to the input prompt and a preset business thinking process in the large language decision model, determining a confidence score corresponding to each organizer, and determining the organizer with the highest confidence score as the target organizer.

[0011] In the second aspect, an embodiment of the present invention also provides a device for recommending a sponsoring unit, the device comprising: an information acquisition module, for acquiring work order information and a preset sponsoring unit set, wherein the work order information comprises: work order content and work order type; a multi-way recall module, for sending the work order content and work order type to a multi-way recommendation model based on traditional language processing, for performing a rough recall process on each sponsoring unit in the preset sponsoring unit set, and for determining a target sponsoring unit set; a decision screening module, for screening the target sponsoring unit set based on a business hybrid thinking chain prompting project and a large language decision model, and for determining a target sponsoring unit, wherein the target sponsoring unit is the most recommended sponsoring unit for solving the problem corresponding to the input work order.

[0012] In a third aspect, an embodiment of the present invention further provides a server, comprising a processor and a memory, wherein the memory stores computer executable instructions that can be executed by the processor, and the processor executes the computer executable instructions to implement any one of the methods provided in the first aspect.

[0013] In a fourth aspect, an embodiment of the present invention further provides a computer-readable storage medium, which stores computer-executable instructions. When the computer-executable instructions are called and executed by a processor, the computer-executable instructions prompt the processor to implement any one of the methods provided in the first aspect.

[0014] The embodiments of the present invention bring the following beneficial effects: The embodiment of the present invention provides a method, device and server for recommending a sponsoring unit. After obtaining work order information and a preset sponsoring unit set, the method sends the work order content and work order type to a multi-channel recommendation model based on traditional language processing, performs a rough recall process on each sponsoring unit in the preset sponsoring unit set, determines a target sponsoring unit set, and then screens the target sponsoring unit set based on a business hybrid thinking chain prompting project and a large language decision model to determine the target sponsoring unit. The embodiment of the present invention can build prior knowledge of sponsoring units through multi-channel recall of business understanding, and build a decision-making thinking chain by summarizing business analysis steps, so that the large model can better understand the decision-making process step by step, make decisions step by step in the process, and finally give a sponsoring unit, thereby significantly improving the accuracy of sponsoring unit recommendation.

[0015] Other features and advantages of the present invention will be described in the following description, and partly become apparent from the description, or understood by practicing the present invention. The purpose and other advantages of the present invention are realized and obtained by the structures particularly pointed out in the description, claims and drawings.

[0016] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, preferred embodiments are given below and described in detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the specific implementation methods of the present invention or the technical solutions in the prior art, the drawings required for use in the specific implementation methods or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some implementation methods of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0018] Figure 1 A flowchart of a recommendation method provided by an organizing unit according to an embodiment of the present invention; Figure 2 A schematic diagram of a specific flow chart of a recommendation method for an organizing unit provided in an embodiment of the present invention; Figure 3 A schematic diagram of the structure of a multi-channel recommendation model provided by an embodiment of the present invention; Figure 4 A schematic diagram of a structure of a recommendation device provided by an organizer according to an embodiment of the present invention; Figure 5 A schematic diagram of the structure of a server provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0019] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution of the present invention will be clearly and completely described in combination with the embodiments below. Obviously, the described embodiments are 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.

[0020] At present, the government service hotline is a service platform established to strengthen interaction with citizens, deepen ties with the people, and effectively solve citizens' life problems. It not only provides citizens with a convenient way to express suggestions and opinions to various departments, make complaints and reports, but also uses the telephone network to build a round-the-clock communication bridge. Among them, building work orders for citizen hotline content and setting accurate labels are crucial to improving service efficiency and quality. Relevant technologies propose that in the process of building work orders for citizen hotline content and setting accurate labels, so as to determine the undertaking unit to solve the work order problem, the undertaking unit can be recommended through traditional classifier training and similarity calculation. However, the undertaking unit recommendation scheme lacks the combination of business and large models in the government field, cannot solve the long tail problem, and the accuracy of the undertaking unit recommendation is low.

[0021] Based on this, the recommendation method, device and server for the hosting unit provided by the present invention explore the optimization plan of the hosting unit in the field of government affairs, combine the traditional natural language processing (NLP) multi-channel recommendation idea and large model technology, first use the NLP recommendation idea as a coarse recall mechanism, and optimize the shortcomings of the traditional recall circuit through business thinking chain prompts, and further use the thinking chain prompts to feed the large model to make its effect more significant. This method effectively makes up for the decision-making based only on traditional historical data reliance, and optimizes the long-tail problem of ordinary machine learning samples after balancing. By organizing the thinking chain to allow the large model to assist in decision-making, and significantly correcting the defects in the incorrect hosting order process in the multi-channel recall, the accuracy of the recommendation of the hosting unit can be significantly improved.

[0022] See also Figure 1 The flowchart of a recommendation method of an organizer is shown, and the method mainly includes the following steps S102 to S106: Step S102, obtain work order information and a set of preset undertaking units, wherein the work order information includes: work order content and work order type. In one embodiment, the work order refers to text information of complaint information or consultation phone number, etc. After receiving the work order information, it is necessary to match the optimal government affairs undertaking unit to solve the work order problem.

[0023] Step S104, sending the work order content and work order type to a multi-way recommendation model based on traditional language processing, performing a rough recall process on each of the preset set of contractors, and determining a target set of contractors. In one embodiment, after the work order content is determined, potential contractors are recalled through a variety of different recall strategies. The number of contractors recalled by any recall strategy may be one or more. When there is no contractor in the preset set of contractors that meets the recall conditions of the recall strategy, the number of contractors recalled by the strategy may also be 0.

[0024] Step S106, based on the business hybrid thinking chain prompt project and the big language decision model, the target undertaking unit set is screened and processed to determine the target undertaking unit, wherein the target undertaking unit is the most recommended undertaking unit for solving the problem corresponding to the input work order. In one embodiment, based on the target undertaking unit set recommended by each call-up circuit and the input prompt determined based on the business hybrid thinking chain prompt project, the Qianwen 32b model (i.e., the big language decision model) is used as an intelligent decision maker to correctly screen and correct the incorrect ones in the a priori set of target undertaking units, so as to determine the target undertaking unit.

[0025] The method for recommending an organizing unit provided by the embodiment of the present invention can significantly improve the accuracy of the organizing unit recommendation.

[0026] See also Figure 2 The specific flow chart of a method for recommending a sponsor is shown in FIG. 1 . The embodiment of the present invention also provides an implementation method for sponsor recommendation, and the details are as follows (A) to (C): (A) After obtaining the work order content and work order type, a multi-channel recall process is performed through a multi-channel recommendation model to determine the target contractor set. For details about the multi-channel recommendation model, see Figure 3 A structural diagram of a multi-channel recommendation model is shown. The various recall strategies in the multi-channel recommendation model include: business rule recall strategy, classifier recall strategy, point problem classification recall strategy, event recall strategy, organizer responsibility recall strategy and similar work order content recall strategy. That is to say, after determining the work order content and work order type, the various recall strategies in the multi-channel recommendation model are used to perform rough recall processing on the organizers, determine the recommended organizers corresponding to each recall strategy, and determine the set of recommended organizers as the target organizer set. For details, see (a) to (f) below: (a) By presetting standardized business rules, the categories to which the work order content belongs are matched, and the recommended undertaking unit corresponding to the business rule recall strategy is determined. Specifically, a set of standardized business rules can be summarized through the experience of dispatching orders in the field of business. Based on the category to which the work order content belongs, the rules of which undertaking unit the work order should be dispatched are preliminarily calculated and recalled, and the rules that meet the current business are recommended for output, that is, the undertaking unit corresponding to the work order problem.

[0027] (b) Based on the classifier of the number of recalls of the historical organizers, the work order content is matched to determine the recommended organizer corresponding to the classifier recall strategy. Specifically, statistical analysis can be performed on the data of the past two years to view the statistical distribution of the organizers, and samples can be screened for the data of the past two years. Based on the statistical analysis, each organizer is extracted according to the frequency quantity. Among them, because the data has a long tail problem, the long tail data needs to be eliminated to keep the model sample balanced. The strategy selects the value threshold of the organizer frequency between 100 and 3000 times. The sample collection method adopts mixed sampling to sample more than 3000 organizers, so as to achieve the best model effect. In addition, the BERT 768-dimensional model used on the algorithm model side is trained as the basic model, so as to recommend the organizer with reference to the semantic information of the past two years.

[0028] (c) Based on the work order type and the point information in the work order content, the work order content is vectorized, and through dot product similarity calculation, the unit with the highest similarity with the work order point information in the preset unit set is determined as the recommended unit corresponding to the point problem classification recall strategy. Specifically, the work order content in the past five years can be vectorized and stored through similarity search based on the point information, location description, and preliminary category and title of the work order content. After that, the vector database is the faiss vector retrieval database, and the dot product calculation is similar. The closest historical unit is returned to obtain the unit that is closest to the work order in terms of problem type and geographical location.

[0029] (d) Events are extracted from historical work orders and an event vector database is constructed to match the organizer corresponding to the work order with the highest similarity in the event vector database and determine it as the recommended organizer corresponding to the event recall strategy. Specifically, events can be extracted from work orders and events are extracted from work orders in the past five years to construct an event vector database. Events can be extracted from new work orders to retrieve similar events and recall the organizer with the most similar events.

[0030] (e) Obtain the portrait of the undertaking unit, and determine the recommended undertaking unit corresponding to the undertaking unit responsibility recall strategy by matching the keywords in the work order content with the keywords in the undertaking unit portrait. That is to say, similarity matching is performed between some keywords in the work order content and the keywords in the undertaking unit portrait, and the most matching undertaking unit is returned. In actual applications, 2 million work order data and 3,000 pieces of unit responsibility data for each final undertaking unit can be selected. The fields used are the main description content of the work order, and finally the model multi-way recall RAG, Qwen 1.5 32B is used.

[0031] (f) The work order content within the preset time threshold is vectorized, and similar recall is performed through text content vectorization and motor calculation to determine the recommended contractor corresponding to the similar work order content recall strategy. Specifically, the work order content of the past five years can be vectorized and stored in the database, and then the text content is vectorized and similar recall is performed through dot product calculation to recall the contractor of the most similar work order content. The dot product calculation formula is: a·b=|a||b|cosθ; In one implementation, candidate organizers = recall loop 1 + recall loop 2 + recall loop 3 + recall loop 4 + recall loop 5 + recall loop 6, an interface is encapsulated for each path, and multi-path organizer selection is achieved through multi-threaded concurrent calling to obtain the final multi-path recall result.

[0032] (B) Generate input prompts corresponding to the target hosting unit set through the business hybrid thinking chain prompt project. Specifically, after marking the source of the target hosting unit set, the business hybrid thinking chain prompt project automatically generates input prompts corresponding to each hosting unit, wherein the input prompts include: background, role, task, main content of the work order and reason for recommending the hosting unit. In one implementation method, the recall result injection is in an ideal state. The input work order data will return to the respective hosting units according to the above six paths, and indicate the source: [{"label source":"classifier recall","label":"**group"}; {"label source":"point, problem classification, title recall","label":"**street"}; {"label source":"event recall","label":"**center"}; {“label source”: “similar recall”, “label”: “** sub-center”}; {“Tag source”: “Business rule recall”, “Tag”: “** sub-center”}; {“Tag Source”:“Unit Responsibility Recall”,“Tag”:“**Sub-Center”},].

[0033] Furthermore, the writing of the prompt includes: Background: There is a work order to be dispatched, and now {6} different agents have given recommendations on the units to be dispatched from different perspectives; Role: You are a senior agent; Task: Please select the three most reasonable hosting units from the recommended hosting units by {6} agents, rank these three hosting units based on rationality and give recommendation values; The main content of the work order and the hosting units and reasons recommended by {6} agents are as follows: Work order content "{}".

[0034] (C) Determine the target organizer from the target organizer set based on the input prompt through the big language decision model. Specifically, according to the input prompt and the preset business thinking process in the big language decision model, perform confidence evaluation on each organizer in the target organizer set, determine the confidence score corresponding to each organizer, and determine the organizer with the highest confidence score as the target organizer. In one embodiment, the business thinking steps include the following (1) to (6): (1) First, determine whether the work order contains any words that endanger social security, the safety of citizens’ lives, or are emotionally agitated. If so, the <110 Command Center> must be added to the organizing unit and ranked first in the recommended list.

[0035] (2) Secondly, pay attention to the specific location mentioned in the work order, confirm the administrative division (district, county, street, community) and surrounding environment to which its geographical location belongs. Analyze whether there has been a record of similar problems at the location, and which unit or units handled the previous problems.

[0036] (3) Review the cases of handling work orders for similar problems that occurred at the same or similar locations in the past. Analyze how the work orders in these cases were classified and handled, and which unit or units were ultimately responsible. Focus on analyzing the handling results of similar problems that occurred at the same or similar locations in the past, and whether the handling results clearly indicate that the problem is the responsibility of other units and the responsible unit is responsible for informing. If this is the case, the corresponding responsible unit can be recommended based on the content of the handling results.

[0037] (4) Check relevant policies, regulations, industry standards and operating procedures, especially whether there are special regulations or guidelines for this point or problem classification. These regulations or guidelines may directly specify the organizer or provide a basis for judgment.

[0038] (5) Review and familiarize yourself with the responsibilities and business areas of each contractor, especially those directly related to the problem location and problem classification. Compare the problem described in the work order with the responsibilities of each unit one by one, and find the most suitable unit or department. Pay special attention to those departments whose responsibilities may overlap or be ambiguous, and determine the most appropriate contractor through in-depth analysis.

[0039] (6) If you encounter special or complex situations, use innovative thinking, consider the nature of the problem and the real needs of the citizens, and flexibly select or recommend the most appropriate organizer. If the problem description involves multiple demands, you can split the incident and select or recommend the corresponding organizer one by one.

[0040] In one embodiment, the large language decision model thinks step by step according to the above steps (1) to (6), and finally combines the nature of the problem and the real needs of the citizens to recommend a reasonable organizer based on the above judgment steps, wherein the output format is: <organizer name 1><recommended value 1>...; the output requirement is: the sum of the recommended values ​​is 100%.

[0041] In practical applications, if there are 6 candidate results r1, r2, ..., rn generated by the recall path and each result has a corresponding confidence score s1, s2, ..., sn, the most correct result can be selected based on the weighted score by the following formula: Rbest=Riwherei=argmaxj(sj); Among them, sj is the score of the j-th result, and argmaxj means selecting the j that maximizes the score sj.

[0042] Furthermore, the data composition can be verified by selecting 200 work order data. Specifically: randomly select 30 work order data from different contractors from the work order data covered by business rules; randomly select 30 work order data from different contractors from the work order data covered by unit responsibilities; select 40 work order data from different contractors from the work orders with complete locations, problem classifications, and titles; and randomly select 100 work order data from different contractors from the overall data. For each piece of data in the verification data, the prompt engineering output is used to obtain three recommended contractors, and the accuracy of the prediction results is evaluated from the perspective of real labels and business. Through the above verification, it can be determined that the solution of the present invention optimizes the overall verification by 4% for the recommended contractor and the long-tail problem compared with the traditional model.

[0043] To sum up, the present invention can combine the traditional multi-channel calculation recall with the big model through the thinking chain, and perform intelligent multi-factor enhancement of multi-channel input, so that the big model can achieve intelligent decision-making. In this way, a significantly novel domain big model decision-making robot is successfully constructed, opening up a new path for intelligent processing in the government field.

[0044] The present invention combines the guidance of business personnel with top-level large-scale models to construct a high-quality data set. By training the 12345 citizen hotline text data, the supervised fine-tuning of the large-scale model is realized, and the model parameters are further superimposed to finally obtain a large-scale model specifically for the 12345 business field. In addition, the idea of ​​a plug-in knowledge base is introduced to perform knowledge plug-ins on domain labels and restrict the output results to prevent the problem of "hallucination" in the output results of large models. The knowledge with domain characteristics is integrated into the large-scale model to realize intelligent generative label recommendation and better meet the needs of application departments.

[0045] Furthermore, in actual applications, by constructing a large-scale domain fine-tuning model and a knowledge label plug-in, candidate domain labels can be recalled from multiple angles, thereby effectively optimizing the existing label processing links. In addition, this method also successfully solves the problem of lack of business characteristics of labels in domain applications, enhances the business attributes of labels, and is more in line with people's usage habits. This not only reduces the workload of labelers, but also improves the performance of the model and the output capacity of the business, bringing substantial improvements to the intelligent processing in the government field.

[0046] For the recommendation method of the sponsor provided in the above embodiment, the present invention provides a recommendation device for the sponsor, see Figure 4 The schematic diagram of the structure of a device recommended by the organizer is shown, and the device includes the following parts: The information acquisition module 402 acquires work order information and a set of preset undertaking units, wherein the work order information includes: work order content and work order type; The multi-channel recall module 404 sends the work order content and work order type to the multi-channel recommendation model based on traditional language processing, performs a rough recall process on each of the pre-set set of undertaking units, and determines the target set of undertaking units; The decision screening module 406 screens the target undertaking unit set based on the business hybrid thinking chain prompting project and the big language decision model to determine the target undertaking unit, wherein the target undertaking unit is the most recommended undertaking unit for solving the problem corresponding to the input work order.

[0047] The recommendation device of the above-mentioned organizing unit provided in the embodiment of the present application can significantly improve the accuracy of the organizing unit's recommendation.

[0048] In one embodiment, when sending the work order content and work order type to a multi-way recommendation model based on traditional language processing, performing a rough recall process on each undertaking unit in a preset undertaking unit set, and determining a target undertaking unit set, the above-mentioned multi-way recall module 404 is also used for: after determining the work order content and work order type, performing a rough recall process on each undertaking unit through each recall strategy in the multi-way recommendation model respectively, determining the recommended undertaking unit corresponding to each recall strategy, and determining the set of each recommended undertaking unit as the target undertaking unit set, wherein the various recall strategies in the multi-way recommendation model include: business rule recall strategy, classifier recall strategy, point problem classification recall strategy, event recall strategy, undertaking unit responsibility recall strategy and similar work order content recall strategy.

[0049] In one embodiment, when performing a rough recall process on each hosting unit through each recall strategy in the multi-way recommendation model and determining the recommended hosting unit corresponding to each recall strategy, the multi-way recall module 404 is also used to: match the category to which the work order content belongs through preset standardized business rules to determine the recommended hosting unit corresponding to the business rule recall strategy; match the work order content with a classifier based on the historical number of hosting unit recalls to determine the recommended hosting unit corresponding to the classifier recall strategy; vectorize the work order content according to the work order type and the point information in the work order content, and determine the hosting unit with the highest similarity with the point information of the work order in the preset hosting unit set as the recommended hosting unit corresponding to the point problem classification recall strategy through dot product similarity calculation.

[0050] In one embodiment, the multi-channel recall module 404 is also used to: extract events for historical work orders, build an event vector database, match the contractor corresponding to the work order with the highest similarity in the event vector database, and determine it as the recommended contractor corresponding to the event recall strategy; obtain a portrait of the contractor, and determine the recommended contractor corresponding to the contractor responsibility recall strategy by matching the keywords in the work order content with the keywords in the contractor portrait for similarity; vectorize the work order content within a preset time threshold, and perform similarity recall through text content vectorization and motor calculation to determine the recommended contractor corresponding to the similar work order content recall strategy.

[0051] In one embodiment, when performing the step of screening the target hosting unit set based on the business hybrid thinking chain prompt project and the big language decision model to determine the target hosting unit, the above-mentioned decision screening module 406 is also used to: generate input prompts corresponding to the target hosting unit set through the business hybrid thinking chain prompt project; and determine the target hosting unit from the target hosting unit set based on the input prompts through the big language decision model.

[0052] In one embodiment, when performing the step of generating input prompts corresponding to the target hosting unit set through the business hybrid thinking chain prompt project, the above-mentioned decision screening module 406 is also used to: after marking the source of the target hosting unit set, automatically generate input prompts corresponding to each hosting unit through the business hybrid thinking chain prompt project, wherein the input prompts include: background, role, task, main content of the work order and reason for recommending the hosting unit.

[0053] In one embodiment, when performing the step of determining the target organizer from the target organizer set based on the input prompt through the big language decision model, the decision screening module 406 is also used to: perform confidence assessment on each organizer in the target organizer set according to the input prompt and the preset business thinking process in the big language decision model, determine the confidence score corresponding to each organizer, and determine the organizer with the highest confidence score as the target organizer.

[0054] The device provided in the embodiment of the present invention has the same implementation principle and technical effects as those of the aforementioned method embodiment. For the sake of brief description, for matters not mentioned in the device embodiment, reference may be made to the corresponding contents in the aforementioned method embodiment.

[0055] An embodiment of the present invention provides a server. Specifically, the server includes a processor and a storage device. The storage device stores a computer program, and when the computer program is executed by the processor, it executes the method described in any one of the above-mentioned embodiments.

[0056] Figure 5A structural diagram of a server provided in an embodiment of the present invention, the server 100 includes: a processor 50, a memory 51, a bus 52 and a communication interface 53, wherein the processor 50, the communication interface 53 and the memory 51 are connected via the bus 52; the processor 50 is used to execute an executable module stored in the memory 51, such as a computer program.

[0057] The memory 51 may include a high-speed random access memory (RAM), and may also include a non-volatile memory, such as at least one disk memory. The communication connection between the system network element and at least one other network element is realized through at least one communication interface 53 (which may be wired or wireless), and the Internet, wide area network, local area network, metropolitan area network, etc. may be used.

[0058] The bus 52 may be an ISA bus, a PCI bus, or an EISA bus, etc. The bus may be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Figure 5 Only one bidirectional arrow is used in the diagram, but this does not mean that there is only one bus or only one type of bus.

[0059] Among them, the memory 51 is used to store programs, and the processor 50 executes the program after receiving the execution instruction. The method executed by the device for flow process definition disclosed in any embodiment of the above-mentioned embodiment of the present invention can be applied to the processor 50 or implemented by the processor 50.

[0060] The processor 50 may be an integrated circuit chip with signal processing capabilities. In the implementation process, each step of the above method can be completed by the hardware integrated logic circuit or software instructions in the processor 50. The above processor 50 can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components. The methods, steps and logic block diagrams disclosed in the embodiments of the present invention can be implemented or executed. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor. The steps of the method disclosed in the embodiments of the present invention can be directly embodied as a hardware decoding processor to be executed, or can be executed by a combination of hardware and software modules in the decoding processor. The software module can be located in a mature storage medium in the field such as a random access memory, a flash memory, a read-only memory, a programmable read-only memory, or an electrically erasable programmable memory, a register, etc. The storage medium is located in the memory 51, and the processor 50 reads the information in the memory 51 and completes the steps of the above method in combination with its hardware.

[0061] The computer program product of the readable storage medium provided in the embodiment of the present invention includes a computer-readable storage medium storing program code, and the instructions included in the program code can be used to execute the methods described in the previous method embodiments. The specific implementation can be referred to the previous method embodiments, which will not be repeated here.

[0062] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions to enable a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the methods described in various embodiments of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc., which can store program codes.

[0063] Finally, it should be noted that the above-described embodiments are only specific implementations of the present invention, which are used to illustrate the technical solutions of the present invention, rather than to limit them. The protection scope of the present invention is not limited thereto. Although the present invention is described in detail with reference to the above-described embodiments, ordinary technicians in the field should understand that any technician familiar with the technical field can still modify the technical solutions recorded in the above-described embodiments within the technical scope disclosed by the present invention, or can easily think of changes, or make equivalent replacements for some of the technical features therein; and these modifications, changes or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention shall be subject to the protection scope of the claims.

Claims

1. A recommendation method for an organizer, characterized in that: The method comprises: Obtaining work order information and a set of preset undertaking units, wherein the work order information includes: work order content and work order type; Sending the work order content and the work order type to a multi-channel recommendation model based on traditional language processing, performing a rough recall process on each of the pre-set organizers set, and determining a target organizer set; Based on the business hybrid thinking chain prompting project and the big language decision model, the target undertaking unit set is screened and processed to determine the target undertaking unit, wherein the target undertaking unit is the most recommended undertaking unit for solving the problem corresponding to the input work order.

2. The recommendation method of the organizer according to claim 1, characterized in that: The step of sending the work order content and the work order type to a multi-channel recommendation model based on traditional language processing, performing a rough recall process on each of the preset set of undertaking units, and determining a target set of undertaking units includes: After determining the work order content and the work order type, each recall strategy in the multi-channel recommendation model is used to perform a rough recall on each of the undertaking units, determine the recommended undertaking units corresponding to each of the recall strategies, and determine the set of the recommended undertaking units as the target undertaking unit set, wherein the recall strategies in the multi-channel recommendation model include: business rule recall strategy, classifier recall strategy, point problem classification recall strategy, event recall strategy, undertaking unit responsibility recall strategy and similar work order content recall strategy.

3. The recommendation method of the organizer according to claim 2, characterized in that: The step of performing rough recall processing on each of the sponsoring units through each recall strategy in the multi-channel recommendation model and determining the recommended sponsoring unit corresponding to each recall strategy includes: By presetting standardized business rules, the categories to which the work order content belongs are matched, and the recommended undertaking unit corresponding to the business rule recall strategy is determined; Based on the classifier of the number of recalls of historical contractors, the work order content is matched to determine the recommended contractor corresponding to the classifier recall strategy; According to the work order type and the point information in the work order content, the work order content is vectorized, and through dot product similarity calculation, the organizer with the highest similarity with the point information of the work order in the preset organizer set is determined as the recommended organizer corresponding to the point problem classification recall strategy.

4. The recommendation method of the organizer according to claim 3, characterized in that: The method further comprises: Extract events from historical work orders and build an event vector database to match the organizer corresponding to the work order with the highest similarity in the event vector database and determine it as the recommended organizer corresponding to the event recall strategy; Obtaining a portrait of the undertaking unit, and determining a recommended undertaking unit corresponding to the undertaking unit responsibility recall strategy by matching the keywords in the work order content with the keywords in the undertaking unit portrait for similarity; The work order content within a preset time threshold is vectorized, and similarity recall is performed through text content vectorization and motor calculation to determine the recommended hosting unit corresponding to the similar work order content recall strategy.

5. The method for recommending an organizer according to claim 1, characterized in that: The step of screening the target organizer set based on the business hybrid thinking chain prompting project and the big language decision model to determine the target organizer includes: Generate input prompts corresponding to the target organizer set through the business hybrid thinking chain prompt project; The target sponsor is determined from the target sponsor set based on the input prompt through the large language decision model.

6. The method for recommending by the organizer according to claim 5, characterized in that: The step of generating input prompts corresponding to the target undertaking unit set through the business hybrid thinking chain prompt project includes: After marking the source of the target hosting unit set, the input prompts corresponding to each hosting unit are automatically generated through the business hybrid thinking chain prompt project, wherein the input prompts include: background, role, task, main content of the work order and reason for recommending the hosting unit.

7. The method for recommending by the organizer according to claim 5, characterized in that: The step of determining the target sponsor from the target sponsor set by using the large language decision model based on the input prompt comprises: According to the input prompt and the preset business thinking process in the large language decision model, a confidence assessment is performed on each organizer in the target organizer set, the confidence score corresponding to each organizer is determined, and the organizer with the highest confidence score is determined as the target organizer.

8. A recommendation device for an organizer, characterized in that: The device comprises: An information acquisition module is used to acquire work order information and a set of preset undertaking units, wherein the work order information includes: work order content and work order type; A multi-channel recall module sends the work order content and the work order type to a multi-channel recommendation model based on traditional language processing, performs a rough recall process on each of the pre-set set of undertaking units, and determines a target set of undertaking units; The decision screening module screens the target undertaking unit set based on the business hybrid thinking chain prompting project and the big language decision model to determine the target undertaking unit, wherein the target undertaking unit is the most recommended undertaking unit for solving the problem corresponding to the input work order.

9. A server, characterized in that: The method comprises a processor and a memory, wherein the memory stores computer executable instructions that can be executed by the processor, and the processor executes the computer executable instructions to implement the method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer-executable instructions, and when the computer-executable instructions are called and executed by a processor, the computer-executable instructions prompt the processor to implement the method according to any one of claims 1 to 7.

Images