An intelligent information recommendation method, apparatus, device, and storage medium

By reusing multi-level document parsing results based on content addressing and cascading fill-in recall, combined with semantic fusion ranking based on adaptive weights, the problems of high parsing and extraction costs, insufficient candidate coverage and accuracy in existing technologies are solved, thereby improving cost-effectiveness and user experience.

CN123115403APending Publication Date: 2026-11-17SHIQU INTERACTIVE (BEIJING) TECH CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611320270.X
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-08-28
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

In existing technologies, the parsing results are not reused at the content level, the extraction intensity is "one-size-fits-all", the candidate pool replacement mechanism is missing, the ranking weight is fixed and lacks feedback loop, it is difficult to balance the freshness of the results and the computational cost, and the grading strategy is rigid in weak matching scenarios, resulting in high parsing and extraction costs and insufficient candidate coverage and accuracy.

Method used

We employ a content-address-based multi-level document parsing result reuse method, a cascaded fill-in recall strategy, and an adaptive weighted semantic fusion sorting method. By combining global snapshot caching with local real-time field hybrid output, we reduce parsing costs, control extraction costs, and balance candidate coverage and accuracy.

Benefits of technology

It achieves a significant reduction in parsing costs, controllable extraction costs, a balance between candidate coverage and accuracy, agile and adjustable sorting strategies, and a stable user experience, meeting the needs of actual work.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123115403A_ABST
    Figure CN123115403A_ABST
Patent Text Reader

Abstract

This application discloses an intelligent information recommendation method, apparatus, device, and storage medium, relating to the field of information extraction technology. The method includes: extracting information from an original file to obtain corresponding basic information fields and tacit knowledge fields; querying a preset project candidate library to obtain a set of similar project candidates based on the basic information fields and tacit knowledge fields; evaluating the overall semantic relevance between each similar project in the candidate set and the basic information fields and tacit knowledge fields, and ranking each similar project in the candidate set; and obtaining the information fields of each similar project based on the ranked candidate set. This application reduces parsing costs, controls extraction costs, balances candidate coverage and accuracy, overcomes the shortcomings of existing technologies, and meets practical work needs.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This application relates to the field of information extraction technology, specifically to methods, apparatus, devices, and storage media for intelligent information recommendation. Background Technology

[0002] Currently, intelligent recommendation of business opportunities / leads to the execution team (or historical reference cases) is a common scenario in enterprise internal knowledge management and resource scheduling systems. Existing related systems generally include: OCR / document understanding and parsing of attachment documents, information extraction based on large models, and candidate recall and ranking based on structured tags or semantic similarity. Among them, the ranking stage often uses a linear weighted formula with fixed weights or a single similarity model. For scenarios with insufficient candidate numbers, two simple strategies are usually adopted: "relaxing all filtering conditions at once" or directly returning insufficient results.

[0003] Therefore, to meet practical needs, an intelligent information recommendation technology is proposed. Summary of the Invention

[0004] To address the shortcomings of existing technologies, the purpose of this application is to provide an intelligent information recommendation method, apparatus, device, and storage medium that reduces parsing costs, controls extraction costs, balances candidate coverage and accuracy, overcomes the deficiencies of existing technologies, and meets practical work needs.

[0005] To achieve the above objectives, the technical solution adopted in this application is as follows: Firstly, this application provides an intelligent information recommendation method, the method comprising the following steps: Based on the original file, information is extracted to obtain the corresponding basic information fields and tacit knowledge fields. Based on the basic information fields and the implicit knowledge fields, a set of similar project candidates is obtained by querying the preset project candidate library; Evaluate the overall semantic relevance between each similar item in the candidate set of similar items and the basic information field and the tacit knowledge field, and rank each similar item in the candidate set of similar items; Based on the sorted candidate set of similar projects, information fields for each of the similar projects are obtained.

[0006] Based on the above technical solution, the method further includes the following steps: The basic information fields and implicit knowledge fields of the original file are synchronously updated to the similar project candidate set.

[0007] Based on the above technical solution, the basic information fields include project name, client name, industry, brand name, business type, project amount, execution cycle, responsible team, project region, project status, business line, cooperation mode, multi-value project tags, and key milestone nodes.

[0008] Based on the above technical solutions, the tacit knowledge fields include consumer insights, target audience profiles, marketing objectives, core creative themes, communication strategies, media mix, detailed pricing structure, cost composition percentages, key success factors, project risk points, customer decision-making chains, customer preferences, collaborator profiles, execution challenges, post-mortem conclusions, performance achievement, conversion funnel characteristics, content tone, competitive landscape, budget allocation logic, creative methodology, data asset accumulation, compliance considerations, customer satisfaction signals, and renewal / repurchase tendencies.

[0009] Secondly, this application provides an intelligent information recommendation device, the device comprising: The information extraction module is used to extract information from the original file to obtain the corresponding basic information fields and tacit knowledge fields. The candidate set query module is used to query and obtain a similar project candidate set from a preset project candidate library based on the basic information field and the implicit knowledge field. The similar item ranking module is used to evaluate the overall semantic relevance between each similar item in the candidate set of similar items and the basic information field and the implicit knowledge field, and to rank each similar item in the candidate set of similar items. The project information extraction module is used to obtain the information fields of each of the similar projects based on the sorted candidate set of similar projects.

[0010] Based on the above technical solution, the device further includes: The candidate set update module is used to synchronously update the basic information fields and implicit knowledge fields of the original file to the similar project candidate set.

[0011] Based on the above technical solution, the basic information fields include project name, client name, industry, brand name, business type, project amount, execution cycle, responsible team, project region, project status, business line, cooperation mode, multi-value project tags, and key milestone nodes.

[0012] Based on the above technical solutions, the tacit knowledge fields include consumer insights, target audience profiles, marketing objectives, core creative themes, communication strategies, media mix, detailed pricing structure, cost composition percentages, key success factors, project risk points, customer decision-making chains, customer preferences, collaborator profiles, execution challenges, post-mortem conclusions, performance achievement, conversion funnel characteristics, content tone, competitive landscape, budget allocation logic, creative methodology, data asset accumulation, compliance considerations, customer satisfaction signals, and renewal / repurchase tendencies.

[0013] Thirdly, this application provides an apparatus including a processor and a memory, the memory being coupled to the processor, the memory storing instructions that, when executed by the processor, cause the apparatus to perform the steps of the method mentioned in the first aspect.

[0014] Fourthly, this application provides a medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method mentioned in the first aspect.

[0015] Compared with the prior art, the advantages of this application are: This application reduces analysis costs, controls extraction costs, balances candidate coverage and accuracy, overcomes the shortcomings of existing technologies, and meets practical work needs.

[0016] This application significantly reduces parsing costs: theoretically, the same document only needs to be parsed once, regardless of how many business objects or processes repeatedly reference it. This significantly reduces the number of calls and costs to external document understanding services, and also provides disaster recovery and reuse capabilities for parsing results in scenarios where the original file is missing.

[0017] In this application, the extraction cost is controllable and easily expandable: only objects with high-value attachments are subject to the high cost of multi-document comprehensive extraction, and implicit knowledge extraction extension bits are reserved without affecting the existing data structure, thus balancing extraction depth and computing power cost.

[0018] In this application, candidate coverage and accuracy are balanced: the cascaded supplementation mechanism significantly reduces the proportion of scenarios with insufficient candidate granularity coverage while ensuring the priority of the main recall accuracy, and at the same time prevents the decline in recall quality caused by indiscriminate relaxation of conditions.

[0019] The ranking strategy in this application is agile, adjustable, and continuously self-optimizing: the dimension weights can be started and stopped independently, and with the adaptive adjustment mechanism based on implicit feedback, the ranking strategy can respond to the business judgment of "a certain dimension signal deteriorating" without modifying the code or redeploying, and continuously improve the matching accuracy as data accumulates.

[0020] In this application, the user experience is stable: adaptive grading ensures that the displayed groups are not empty and have distinctiveness under any matching strength; the mixed output mechanism of snapshot and real-time fields ensures that most requests respond quickly, while avoiding the obsolescence of key time-sensitive fields. Attached Figure Description

[0021] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0022] Figure 1 This is a flowchart illustrating the steps of the intelligent information recommendation method according to an embodiment of this application. Figure 2 This is a system architecture diagram of the intelligent information recommendation method according to an embodiment of this application; Figure 3 This is a core flowchart of the intelligent information recommendation method in an embodiment of this application; Figure 4 This is a structural block diagram of the intelligent information recommendation device according to an embodiment of this application. Detailed Implementation

[0023] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0024] The embodiments of this application will be further described in detail below with reference to the accompanying drawings.

[0025] This application provides an intelligent information recommendation method, apparatus, device, and storage medium, which reduces parsing costs, controls extraction costs, balances candidate coverage and accuracy, overcomes the shortcomings of the prior art, and meets practical work needs.

[0026] To achieve the aforementioned technical effects, the overall concept of this application is as follows: An intelligent information recommendation method, comprising the following steps: S1. Based on the original file, extract information to obtain the corresponding basic information fields and tacit knowledge fields; S2. Based on the basic information fields and implicit knowledge fields, a set of similar project candidates is obtained by querying the preset project candidate library; S3. Evaluate the overall semantic relevance between each similar item in the candidate set of similar items and the basic information field and the implicit knowledge field, and rank each similar item in the candidate set of similar items. S4. Based on the sorted candidate set of similar projects, obtain the information fields of each similar project.

[0027] The embodiments of this application will be further described in detail below with reference to the accompanying drawings.

[0028] Firstly, see [the following] Figures 1-3 As shown in the figure, this application provides an intelligent information recommendation method, which includes the following steps: S1. Based on the original file, extract information to obtain the corresponding basic information fields and tacit knowledge fields; S2. Based on the basic information fields and implicit knowledge fields, a set of similar project candidates is obtained by querying the preset project candidate library; S3. Evaluate the overall semantic relevance between each similar item in the candidate set of similar items and the basic information field and the implicit knowledge field, and rank each similar item in the candidate set of similar items. S4. Based on the sorted candidate set of similar projects, obtain the information fields of each similar project.

[0029] It should be noted that the existing technology has the following technical defects: The parsing results are not reused at the content level: The same attachment document (contract, plan, case closure report, etc.) is repeatedly submitted to external OCR / document understanding services for parsing in different business objects and different processing stages. The parsing results are strongly bound to the original file that triggered the parsing. Once the original file is deleted due to storage cleanup or other reasons, the previous parsing results become invalid or cannot be retrieved and reused, resulting in repeated parsing and wasted computing power and call costs.

[0030] The extraction process suffers from a "one-size-fits-all" approach and poor scalability: existing information extraction processes typically apply the same level of processing to all business objects. This results in either only shallow, structured extraction with insufficient information depth, or indiscriminately performing costly multi-document comprehensive extraction on all objects (including those without high-value attachments), leading to wasted computing power. Furthermore, there is a lack of standardized extension methods that allow for the smooth integration of deeper extraction capabilities (such as tacit knowledge) without altering the existing output structure.

[0031] Lack of candidate pool replacement mechanism: When the number of candidates (such as executable teams) is insufficient to meet the preset display requirements, the existing system either simply relaxes all recall filtering conditions (leading to the mixing in of a large number of irrelevant candidates and a sharp drop in sorting quality), or directly returns results with insufficient quantity (leading to a narrow range of choices for business personnel). It lacks a cascading recall strategy that "fills in candidates step by step from strong to weak according to signal strength and only absorbs incremental candidates".

[0032] Fixed sorting weights and lack of feedback loop: The weights of each dimension in the sorting formula are usually hard-coded as fixed values ​​in the code. When noise is introduced into a certain dimension due to insufficient accuracy of upstream information extraction, it can only be adjusted by modifying the code and redeploying, resulting in a long response chain. At the same time, once the weights are set, they remain unchanged for a long time and cannot be automatically optimized by utilizing implicit feedback signals such as user browsing and clicks continuously generated during system operation. The sorting accuracy cannot improve itself with the accumulation of data.

[0033] The results show that freshness and computational cost are difficult to balance: the ranking results either have to be calculated entirely in real time (which involves re-executing the complete recall, scoring, semantic re-ranking, large model generation and interpretation steps for each request, resulting in high response latency and high cost), or they have to be cached entirely offline (the results may have deviated from the latest state of the candidates, resulting in insufficient freshness). There is no compromise between the two that balances cost and freshness.

[0034] Rigid tiering strategies in weak matching scenarios: When the overall matching degree between the candidate pool and the query is generally weak (such as all candidate ranking scores being lower than the preset "high matching" absolute threshold), the existing fixed absolute threshold tiering method is prone to causing all candidates to be classified into the same tier or even the lowest tier, causing the front-end display grouping to collapse and lose its differentiation, and failing to provide business personnel with an effective priority reference.

[0035] In this embodiment, the analysis cost is reduced, the extraction cost is controlled, and the candidate coverage and accuracy are balanced, overcoming the shortcomings of the prior art and meeting the needs of actual work.

[0036] Furthermore, the method also includes the following steps: The basic information fields and implicit knowledge fields of the original file are synchronously updated to the similar project candidate set.

[0037] Furthermore, the basic information fields include project name, client name, industry, brand name, business type, project amount, execution period, responsible team, project region, project status, business line, cooperation mode, multi-value project tags, and key milestone nodes.

[0038] Furthermore, the tacit knowledge fields include consumer insights, target audience profiles, marketing objectives, core creative themes, communication strategies, media mix, detailed pricing structure, cost composition percentages, key success factors, project risk points, customer decision-making chain, customer preferences, collaborator profiles, execution challenges, post-mortem conclusions, performance achievement, conversion funnel characteristics, content tone, competitive landscape, budget allocation logic, creative methodology, data asset accumulation, compliance considerations, customer satisfaction signals, and renewal / repurchase tendencies.

[0039] Based on the technical solution of this application embodiment, it aims to provide a pluggable, adaptive, cascaded, and content-level reusable intelligent extraction and matching method, the main logic of which is divided into two links: (a) Offline extraction link (write side): Project documentation → Basic information fields + Implicit knowledge fields; The project documents are reused through content addressing and parsing in module ① and extracted through hierarchical triggering in module ②, solidifying the output of "basic information fields + implicit knowledge fields" and accumulating them into a searchable project candidate library.

[0040] Module ①, a method for reusing multi-level document parsing results based on content addressing: Input to module ①: Each attachment document (such as a plan, a final report, or a quotation) connected to the system.

[0041] Intermediate calculation method for module ①: [Simple field matching / deterministic hashing, not an algorithm] Calculate the SHA-256 hash value of the file content as the cache key; Establish a three-level storage system (local cache / database hash index / read-only reuse interface), and query the index by (data source, business primary key, file name or attachment identifier) ​​to obtain the hash value and hit the cache.

[0042] Output of module ①: The parsing results (meta-information or complete parsing content) support hash-based degradation and reuse when the original file is missing, and the external parsing service is not called repeatedly throughout the process.

[0043] Module 2, Hierarchical Triggered Structured-Multimodal Information Extraction Method: Input for module ②: Structured fields already in the database + raw form / push data + attached documents (high-value types trigger deep extraction).

[0044] Intermediate calculation method for module ②: [Algorithm (Large Language Model Extraction)] The first level forces the extraction of basic structured facts from all objects; the second level triggers multi-document comprehensive extraction on high-value documents to produce in-depth professional information; a third level of "tacit knowledge extraction" is reserved for pluggable expansion (requires meeting minutes, emails, and communication records as input).

[0045] Output of module ②: Basic information fields (14 items: Project Name, Client Name, Industry, Brand Name, Business Type, Project Amount, Execution Period, Responsible Team, Project Region, Project Status, Business Line, Cooperation Mode, Multi-value Project Tags, Key Milestones); Implicit knowledge fields (25 items: consumer insights, target audience profile, marketing objectives, core creative themes, communication strategies, media mix, detailed pricing structure, cost composition, key success factors, project risk points, customer decision chain, customer preferences, collaborator profiles, execution challenges, post-mortem conclusions, results achieved, conversion funnel characteristics, content tone, competitive landscape, budget allocation logic, creative methodology, data asset accumulation, compliance considerations, customer satisfaction signals, renewal / repurchase tendency); Each implicit conclusion is forcibly associated with an evidence anchor point (source document + page number) and a confidence score.

[0046] Specifically, the first list of fixed fields is: Basic Information Fields (14 items, mandatory output from Module ② Level 1), as shown in Table 1 below: Serial Number Basic Information Fields Field Description 1 Project Name The project's unique name in the system 2 Customer Name The client / owner of the project 3 Industry The client's industry classification, such as FMCG, automotive, and finance. 4 Brand Name Specific brands involved in the project 5 Business type Project business nature, such as brand marketing, e-commerce operations, and private domain growth. 6 Project Amount Contract amount or range 7 Execution cycle Project start and end date range 8 Responsible Team The internal team responsible for executing the project 9 Project Area The city or region where the project will be located 10 Project Status In progress, completed, verified 11 Business line The business line to which the project belongs 12 Cooperation Model Agents, direct customers, joint execution 13 Multi-value item tags Multiple-selectable item tag collection 14 Key milestones Key Time Nodes for the Project

[0047] Table 1 Specifically, the second list of fixed fields includes: tacit knowledge fields (25 items, output from Module ② Level 2 / 3), as shown in Table 2 below: Serial Number Implicit knowledge fields Field Description Anchor of evidence requirements Confidence requirement 1 Consumer Insights Insights into the needs, pain points, and behaviors of the target audience The source document and page number must be linked. Give a score of 0–1 2 Target audience profile Audience demographics, interests, and consumption characteristics The source document and page number must be linked. Give a score of 0–1 3 Marketing objectives Four types of objectives: user acquisition, user activation, conversion, and brand exposure. The source document and page number must be linked. Give a score of 0–1 4 Core Creative Theme Core creative concept or theme of the project The source document and page number must be linked. Give a score of 0–1 5 Communication strategy Overall dissemination rhythm and tactics The source document and page number must be linked. Give a score of 0–1 6 Media combination Media channel combination used The source document and page number must be linked. Give a score of 0–1 7 Price list details Items and Structure of the Quotation The source document and page number must be linked. Give a score of 0–1 8 Cost composition percentage Distribution of cost items by percentage The source document and page number must be linked. Give a score of 0–1 9 Key success factors Key drivers of project success The source document and page number must be linked. Give a score of 0–1 10 Project risk points Risks and problems exposed during implementation The source document and page number must be linked. Give a score of 0–1 11 Customer decision chain Key decision-makers and decision-making paths on the customer side The source document and page number must be linked. Give a score of 0–1 12 Customer preferences Customer preferences in style, tone, and suppliers The source document and page number must be linked. Give a score of 0–1 13 Collaborator Profile Characteristics of service providers in the three categories of design, media, and technology The source document and page number must be linked. Give a score of 0–1 14 Implementation difficulties Difficulties encountered during execution The source document and page number must be linked. Give a score of 0–1 15 Review Conclusion Lessons learned from project debriefing The source document and page number must be linked. Give a score of 0–1 16 Results achieved KPI and target achievement rate The source document and page number must be linked. Give a score of 0–1 17 Characteristics of conversion funnel Characteristics of each layer of the conversion funnel The source document and page number must be linked. Give a score of 0–1 18 Content tone The tone and style of the content (humorous, serious, warm). The source document and page number must be linked. Give a score of 0–1 19 Competitive Landscape Competitive environment of the project The source document and page number must be linked. Give a score of 0–1 20 Budget allocation logic Budget allocation among modules The source document and page number must be linked. Give a score of 0–1 21 Creative Methodology Creative methods or frameworks used The source document and page number must be linked. Give a score of 0–1 22 Data asset accumulation Reusable data accumulated from the project The source document and page number must be linked. Give a score of 0–1 23 Compliance Considerations Compliance requirements to be aware of The source document and page number must be linked. Give a score of 0–1 24 Customer satisfaction signals Signals reflecting customer satisfaction The source document and page number must be linked. Give a score of 0–1 25 Renewal / Repurchase Tendency Customers' tendency to renew or make repeat purchases The source document and page number must be linked. Give a score of 0–1

[0048] Table 2 Note: Each tacit knowledge field must include an "evidence anchor (source document + page number)" and a "confidence score (0–1)" to ensure traceability and allow for sorting by confidence level.

[0049] (ii) Online query chain (query side): requirement brief → similar items (information fields + attachments); After the basic information of the requirements brief is extracted by the large model, it is sequentially processed through module ③ recall, module ④ sorting, module ⑤ grading, and module ⑥ output to obtain similar items.

[0050] Module ③, Multi-level Cascaded Filling Recall Method When the Number of Candidates is Insufficient: Input for module ③: Query the basic information extracted from the brief by the big model (industry, brand, business type, amount range, execution cycle, responsible team).

[0051] Intermediate calculation method for module ③: [Hybrid] Establish 3 independent recall channels: ①Structured tag exact matching (with tiered rate limiting, prioritizing completed / verified categories); ② Semantic vector approximate nearest neighbor retrieval (global or specified dimension); ③ Multi-valued label union matching. When the number of candidate granularities does not reach the threshold, the cascaded filling is performed in descending order of signal strength (semantic nearest neighbor → independent business type → most recently active history). Each level only accepts candidates that have not appeared before, thus avoiding duplication and preventing true expansion.

[0052] Output of module ③: A candidate set of similar projects (candidate granularity such as team / case), including their basic information fields and attachment references.

[0053] Module 4, a pluggable, adaptive weighted structured-semantic fusion ranking method: Input for module ④: The candidate set and query context produced by module ③.

[0054] Intermediate calculation method for module ④: [Algorithm] Calculate the structured matching score for each candidate from multiple dimensions (Jaccard similarity, hierarchical matching, interval proximity, and different algorithms can be used for each dimension), and introduce a semantic reordering model to calculate the overall semantic relevance; Final score R = Σ(w i ·s i )+w_rerank·s_rerank; Weight w for each dimension i Both the reranking weight w_rerank and the reranking weight are on / off switches that can be started and stopped independently during runtime; Automatically adjust permissions periodically based on implicit user behavior feedback (only recording the behavior of the entity that initiated the request, isolating third-party behavior).

[0055] Output of module ④: Normalized percentage scores, with candidates sorted from highest to lowest score.

[0056] Module 5, Relative Ranking Adaptive Tiering Method for Weak Matching Scenarios in Candidate Pools: Input for module ⑤: The highest ranking score in this candidate pool.

[0057] Intermediate calculation method for module ⑤:

Simple Threshold Rules

[0058] Output of module ⑤: Display groups that are not empty and have distinctness under any matching strength.

[0059] Module 6, Method for Consistent Output of Global Snapshot Cache and Local Real-Time Fields: Input for module ⑥: Module 4: Real-time query of sorted results and time-sensitive fields.

[0060] Intermediate calculation method of module ⑥:

Simple Storage + Field Overwrite

[0061] Output of Module 6: Information fields of similar projects + attachment references (mixed output of snapshot and real-time fields).

[0062] Based on the above, the technical advantages of the technical solution of this application embodiment are as follows: Module ① Multi-level document parsing result reuse method based on content addressing (reuse across objects and processes, parsing is done only once, and the original file can still be reused even if it is missing).

[0063] Module ③ Multi-level cascaded supplementation recall method when the number of candidates is insufficient (3-channel independent recall + strong to weak cascaded supplementation to ensure candidate coverage).

[0064] Module ④ Pluggable, adaptive weighted structured-semantic fusion ranking method (R=Σ(w i ·s i ()+w_rerank·s_rerank, with independent start and stop of dimension weights and self-adjustment based on feedback, which is the core innovation).

[0065] Secondly, see Figure 4 As shown in the figure, this application provides an intelligent information recommendation device, which includes: The information extraction module is used to extract information from the original file to obtain the corresponding basic information fields and tacit knowledge fields. The candidate set query module is used to query and obtain a similar project candidate set from a preset project candidate library based on the basic information field and the implicit knowledge field. The similar item ranking module is used to evaluate the overall semantic relevance between each similar item in the candidate set of similar items and the basic information field and the implicit knowledge field, and to rank each similar item in the candidate set of similar items. The project information extraction module is used to obtain the information fields of each of the similar projects based on the sorted candidate set of similar projects.

[0066] In this embodiment, the analysis cost is reduced, the extraction cost is controlled, and the candidate coverage and accuracy are balanced, overcoming the shortcomings of the prior art and meeting the needs of actual work.

[0067] Furthermore, the device also includes: The candidate set update module is used to synchronously update the basic information fields and implicit knowledge fields of the original file to the similar project candidate set.

[0068] Furthermore, the basic information fields include project name, client name, industry, brand name, business type, project amount, execution period, responsible team, project region, project status, business line, cooperation mode, multi-value project tags, and key milestone nodes.

[0069] Furthermore, the tacit knowledge fields include consumer insights, target audience profiles, marketing objectives, core creative themes, communication strategies, media mix, detailed pricing structure, cost composition percentages, key success factors, project risk points, customer decision-making chain, customer preferences, collaborator profiles, execution challenges, post-mortem conclusions, performance achievement, conversion funnel characteristics, content tone, competitive landscape, budget allocation logic, creative methodology, data asset accumulation, compliance considerations, customer satisfaction signals, and renewal / repurchase tendencies.

[0070] In summary, the intelligent information recommendation device provided in this application embodiment has the same technical principle as the intelligent information recommendation method provided in the first aspect in terms of technical problems, technical solutions and technical effects, so it will not be described again here.

[0071] Thirdly, embodiments of this application provide a device including a processor and a memory, the memory being coupled to the processor, the memory storing instructions that, when executed by the processor, cause the device to perform the steps of the method mentioned in the first aspect. Fourthly, embodiments of this application provide a medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method mentioned in the first aspect.

[0072] In the description of this application, it should be noted that the terms "upper," "lower," etc., indicating the orientation or positional relationship are based on the orientation or positional relationship shown in the accompanying drawings, and are only for the convenience of describing this application and simplifying the description, and do not indicate or imply that the system or component referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of this application. Unless otherwise expressly specified and limited, the terms "installed," "connected," and "linked" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; they can refer to the internal communication between two components. For those skilled in the art, the specific meaning of the above terms in this application can be understood according to the specific circumstances.

[0073] It should be noted that in this application, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0074] The above description is merely a specific embodiment of this application, enabling those skilled in the art to understand or implement this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features claimed herein.

Claims

1. An intelligent information recommendation method, characterized in that, The method includes the following steps: Based on the original file, information is extracted to obtain the corresponding basic information fields and tacit knowledge fields. Based on the basic information fields and the implicit knowledge fields, a set of similar project candidates is obtained by querying the preset project candidate library; Evaluate the overall semantic relevance between each similar item in the candidate set of similar items and the basic information field and the tacit knowledge field, and rank each similar item in the candidate set of similar items; Based on the sorted candidate set of similar projects, information fields for each of the similar projects are obtained.

2. The intelligent information recommendation method as described in claim 1, characterized in that, The method further includes the following steps: The basic information fields and implicit knowledge fields of the original file are synchronously updated to the similar project candidate set.

3. The intelligent information recommendation method as described in claim 1, characterized in that: The basic information fields include project name, client name, industry, brand name, business type, project amount, execution period, responsible team, project region, project status, business line, cooperation mode, multi-value project tags, and key milestone nodes.

4. The intelligent information recommendation method as described in claim 1, characterized in that: The tacit knowledge fields include consumer insights, target audience profiles, marketing objectives, core creative themes, communication strategies, media mix, detailed pricing structure, cost composition, key success factors, project risk points, customer decision-making chain, customer preferences, collaborator profiles, execution challenges, post-mortem conclusions, performance achievement, conversion funnel characteristics, content tone, competitive landscape, budget allocation logic, creative methodology, data asset accumulation, compliance considerations, customer satisfaction signals, and renewal / repurchase tendencies.

5. An intelligent information recommendation device, characterized in that, The device includes: The information extraction module is used to extract information from the original file to obtain the corresponding basic information fields and tacit knowledge fields. The candidate set query module is used to query and obtain a similar project candidate set from a preset project candidate library based on the basic information field and the implicit knowledge field. The similar item ranking module is used to evaluate the overall semantic relevance between each similar item in the candidate set of similar items and the basic information field and the implicit knowledge field, and to rank each similar item in the candidate set of similar items. The project information extraction module is used to obtain the information fields of each of the similar projects based on the sorted candidate set of similar projects.

6. The intelligent information recommendation device as described in claim 5, characterized in that, The device further includes: The candidate set update module is used to synchronously update the basic information fields and implicit knowledge fields of the original file to the similar project candidate set.

7. The intelligent information recommendation device as described in claim 5, characterized in that: The basic information fields include project name, client name, industry, brand name, business type, project amount, execution period, responsible team, project region, project status, business line, cooperation mode, multi-value project tags, and key milestone nodes.

8. The intelligent information recommendation device as described in claim 5, characterized in that: The tacit knowledge fields include consumer insights, target audience profiles, marketing objectives, core creative themes, communication strategies, media mix, detailed pricing structure, cost composition, key success factors, project risk points, customer decision-making chain, customer preferences, collaborator profiles, execution challenges, post-mortem conclusions, performance achievement, conversion funnel characteristics, content tone, competitive landscape, budget allocation logic, creative methodology, data asset accumulation, compliance considerations, customer satisfaction signals, and renewal / repurchase tendencies.

9. A device, characterized in that, The device includes a processor and a memory, the memory being coupled to the processor, the memory storing instructions that, when executed by the processor, cause the device to perform the steps of the method according to any one of claims 1 to 4.

10. A medium having a computer program stored thereon, characterized in that, When executed by a processor, the program implements the steps of the method described in any one of claims 1 to 4.