CREATING A VERIFIED RECOMMENDATION PROTOCOL BASED ON PROFILE, RISK STATUS AND HYBRID INFORMATION ACCESS. COMPUTERIZED SYSTEMS AND METHODS
Patent Information
- Application Number
- TR202613842
- Authority / Receiving Office
- TR · TR
- Patent Type
- Applications
- Current Assignee / Owner
- Filing Date
- 2026-08-14
- Publication Date
- 2026-09-21
Smart Images
Figure 00000030_0000
Abstract
Description
1 TARIFF PROFILE, RISK STATUS, HYBRID INFORMATION ACCESS, AND MULTI-STAGE. VERIFIED PROPOSAL BASED ON VERIFICATION MECHANISMS COMPUTERIZED SYSTEM AND METHOD THAT CREATED THE PROTOCOL Technical Section: 5 The invention relates to computer-applied systems, artificial intelligence-assisted data processing, and natural language processing. user profiling, dynamic profile management, risk assessment systems, hybrid information access, Information graph-based data management, vector database technologies, evidence grading mechanisms, information verification systems, hallucination detection mechanisms, source consistency analysis, confidence score calculation algorithms, rule-based 10 validation engines, decision support systems, and computer-aided recommendation generation. These technologies relate to technical fields; in particular, user profile data, contextual By evaluating data, risk status, and user query together, from different information sources... processing the obtained data using hybrid information retrieval methods, the level of evidence for the obtained information, 15 It creates a dynamic confidence score for each recommendation based on the verification process, and is contrary to the truth. By identifying and filtering content, it provides reliable, traceable, verifiable, and personalized information. It relates to computer-based systems and methods that create a proposed protocol. State of the Art: Computer-based decision support systems, AI-powered recommendation systems, natural language 20 Processing technologies, information access systems, and personalized recommendations based on user profile. mechanisms; health informatics, nutrition, personal care, cosmetics, natural products, e-commerce and It is widely used in many similar technical fields, especially large language models. (Large Language Models - LLM), Retrieval-Augmented Generation - 2 RAG), knowledge graphs, vector database technologies, and other artificial intelligence Analysis of user queries generated in natural language thanks to information processing-based approaches It has become possible to conduct research and generate contextual suggestions. In current systems, queries submitted by the user, either verbally or in writing, are processed using natural language processing. The query is analyzed using various techniques, key concepts related to the query are identified, and 5 Suggestions are made to the user using content obtained from different information sources. These systems are being developed. Some systems use semantic vector search methods, while others use information-based methods. Graph-based relational data structures are used, and these approaches are employed in advanced systems. They can be implemented together. However, when existing technical solutions are examined, the proposal The creation process largely depends on the statistical predictive ability of the generative artificial intelligence model. It appears to be based on... While generative AI models can generate natural and consistent text, the generated text... the accuracy, reliability, timeliness, and verifiability of the information alone This cannot be guaranteed, especially for up-to-date information not included in the model's training data. the use of conflicting information sources or unverified data 15 content that is untrue, incomplete, contradictory, or not supported by evidence, if processed It can be produced. This phenomenon, called "hallucination" in the literature, is a suggestion. This significantly reduces the reliability of their systems. In addition, what information sources are used to base the recommendations made within the existing systems? Level of scientific or technical evidence from the sources used, currency of the sources, version 20 The information and its consistency with each other are often not evaluated systematically. If information from different sources contradicts each other, which information should take precedence? The lack of a technical decision-making mechanism regarding how the proposed solutions will be used This can negatively affect its accuracy. 3 In a significant number of existing systems, user profiles and query data are independent of each other. This includes the user's age, biological sex, and pregnancy or breastfeeding status. chronic illnesses, allergies, regular medication use, past usage habits, preferences and similar profile data are mostly only a part of natural language input. This information is being evaluated and is dynamically updated by the system. 5 It is not converted into structured profile data. Therefore, it changes over time. User characteristics are not adequately reflected in the recommendation process and user profile. It is not possible to completely prevent the creation of dissenting proposals. On the other hand, semantic search is used in existing AI-based recommendation systems. methods that can successfully identify semantic similarities in natural language queries 10 However; drug interactions, contraindications, age groups, pregnancy status, biological sex, Allergies, active ingredient repetitions, product compatibility, and similar specific technical regulations alone It is unable to evaluate. In contrast, only information graph or rule-based systems can be used. However, it is unable to adequately interpret the semantic diversity in natural language queries, different It is unable to adapt to user expression styles. 15 In addition, in existing systems, candidate data sets are scored according to their reliability. ranking, elimination based on safety rules, and a quantitative confidence score for each proposal. Calculations are often not feasible. Instead, suggestions are directly productive. This is done by transferring it to a model, and numerical data is generated regarding the extent to which the proposal is reliable. or a measurable evaluation cannot be presented to the user. 20 In addition, candidate data sets are systematically analyzed before the proposal generation process. It is not passed through security filters, and independent verification is required after the proposal is created. Content verification mechanisms, evidence consistency analysis, false content 4 It appears that processes such as identification and source verification are not performed together. This This situation can lead to technical errors reaching the end user. Another significant shortcoming of current systems is that user feedback is not directly productive. The ability to integrate artificial intelligence models into learning processes and the accuracy of this feedback. or insufficient oversight in terms of expert approval. This situation is incorrect, incomplete or 5 This can cause manipulative user input to negatively affect system behavior. It is possible. In conclusion, the AI-based recommendation systems currently used in technology; user profile, hybrid information access that integrates contextual data and risk assessment. evidence grading, source contradiction analysis, timeliness management, trust score calculation, 10 It uses, along with, false content detection and multi-stage verification mechanisms, Optimizing recommendations based on dynamically updated user profiles and the entire process an integrated computer-based technical solution that records the process in a traceable manner It is unable to provide this. Therefore, a new computer that resolves these technical problems is needed. An applied system and method are needed. 15 Purpose and Technical Benefits of the Invention: The invention incorporates user profile data, contextual data, user queries, and risk status. By considering them together, hybrid information retrieval, information graph-based data management, and vector databases. technologies, evidence rating, source contradiction analysis, currency updates and version management, and much more. reliable, traceable, verifiable and 20 using a combination of phased verification mechanisms a computer-based system and method that creates a personalized recommendation protocol It is related. One of the main aims of the invention is to analyze the user query based solely on its natural language content. Instead of evaluating the user, it uses profile information, past usage data, and preferences. contextual conditions and dynamically updated risk information in structured data structures. by enabling their evaluation together, the user-specific transaction context This involves creating a profile that reflects the user's changing profile characteristics and usage over time. Their behaviors can be reflected in the recommendation generation process, and recommendations can be tailored to the current user profile. Compatibility is being increased. 5 Another objective of the invention is to obtain information regarding a user query from only one source of information. Instead of obtaining at least one vector index, at least one evidence-interaction information graph, and at least one The aim is to enable hybrid information access through the product / content inventory matrix. This results in semantic similarity, structured information relationships, and current product or content. The information can be evaluated together within the same process chain, and more reliable candidate information can be obtained. 10 Clusters can be formed. Another objective of the invention is to propose candidate knowledge sets obtained as a result of hybrid knowledge access. Technical constraints, security regulations, profile compatibility, and risk level are considered before the creation process. preference parameters include level of evidence, currency of sources, and inter-source consistency. The aim is to ensure that it is evaluated. Thus, inappropriate, insufficient evidence level, 15 Candidates who have submitted proposals that are outdated or technically contradictory to other sources Its transfer to the process is restricted. The invention also includes a system of evidence that classifies sources of information according to the level of evidence they possess. It provides a rating mechanism. This mechanism takes into account the quality of the information source, Source type, expert review, clinical or technical guideline nature, systematic review, 20 meta-analysis and similar predefined categories of evidence regarding source reliability assigning an evidence level or evidence score to the relevant information by evaluating the parameters. This ensures that in the proposal generation process, not only semantic similarity but also the results obtained are considered. The evidentiary value of the information obtained is also taken into consideration. 6 Another purpose of the invention is to collect information from different sources on the same subject, product, content, data that contains different results regarding a component, usage condition, or technical specification. The aim is to provide a source for identifying contradictions, a contradiction analysis mechanism. The mechanism in question identifies differences between sources using specific comparison criteria. By analyzing this, the weight, priority, or 5 of conflicting pieces of information in the candidate pool can be determined. can determine the use case and re-verify the relevant information where necessary. It can direct you to the process. The invention also includes a currency assessment system that evaluates the currency and version status of information sources. and provides a version management mechanism. Through this mechanism, the same or similar Different versions of sources containing information can be compared, and the publication or 10 of the sources can be compared. Current and valid information is determined by evaluating update dates, validity status, and version information. This ensures the prioritization of resources. Thus, old, outdated, or less relevant resources are eliminated. The impact of updated information from a new source on the recommendation generation process. is being reduced. Another aim of the invention is to eliminate candidate data sets solely on the basis of suitability. not only, but also scoring and ranking using multiple technical parameters The aim is to ensure compatibility with the user profile and semantics in relation to the query for candidate information sets. relationship, level of evidence, source reliability, source currency, risk alignment, inter-source Points can be assigned based on consistency and other technical criteria determined by the system, and Candidates who score below the specified threshold may be eliminated. 20 The invention also provides for dynamic reliability for each proposed protocol or proposed component that is generated. It enables the calculation of the confidence score. The confidence score is based on the level of evidence from the sources used. Resource currency, consistency between resources, compatibility with user profile, risk validation. The result is the score of the candidate information set, the verifiability of the information in the proposal, and so on. 7 This can be calculated by evaluating the parameters together. Thus, the system... evaluating the reliability of the proposed solutions based on measurable technical values it becomes possible. Another aim of the invention is a generative artificial intelligence model or similar content generation engine. The information and statements contained in the proposal prepared by [Name of party] are used in the preparation of the proposal. This allows comparison with verified data sets and evidence sources used. The aim is to provide a mechanism for detecting false content and checking the consistency of evidence. This mechanism... proposals generated through this medium that are not supported, not found in the sources, or contradict the sources or identifying content that does not meet the verification criteria and the relevant recommendations section It can be recreated or corrected. 10 The invention also allows for the application of different levels of verification depending on the risk score. It provides: User profile, query content, identified components or products, usage. A risk score is calculated by evaluating the conditions and other risk parameters, and the statement The operating mode of the system is determined according to the score. In low-risk situations, it is more... While low verification density is applicable, higher evidence is required in high-risk situations. 15 level, more source validation, tighter constraint control, additional contradiction analysis and Expert evaluation can be applied when necessary. Another purpose of the invention is to analyze the proposed protocol before and after its production. The goal is to ensure that it undergoes at least two separate verification stages, including production. Candidate data sets and their technical parameters in pre-validation 20 When evaluated, the product and content in the final recommendation generated during post-production verification, The components, quantity, route of use, timing, and similar factors are determined by re-evaluating them. Technical rules and validated data sets are used to compare the generative model. Thus, the generative model... 8 the final text he produced was not only formal, but also in terms of content and consistency of evidence. This ensures that it is monitored. The invention also includes a provision that should any errors or technical inconsistencies be detected during verification. Instead of recreating the entire protocol, the error-containing suggestion block or suggestion itself can be identified. By identifying the component, it is possible to reproduce only the relevant section. 5 This reduces unnecessary model calls, computational load, calculation cost, and latency, and This ensures more efficient use of system resources. Another purpose of the invention is to enable the use of the same active ingredient or technically equivalent content in different products. The aim is to provide a control mechanism that detects the possibility of it being seized again. within the scope of the user's existing product and content usage, as well as new candidate products or 10 The contents are compared; recurring active ingredients, potential interactions, and route of use are examined. Unsuitable candidates are identified by evaluating incompatibilities and other identified technical limitations. Elimination is ensured. The invention also allows user-provided feedback to be directly applied to generative artificial intelligence. a control that prevents the model from being transferred to the system in a way that would change its behavior 15 It provides a mechanism for user feedback, primarily for verification and consistency checking. and undergo expert review processes when necessary, only verified or Authorized feedback in the system's profile, rule, or suggestion optimization processes Its use is permitted. Another purpose of the invention is to gather verified user feedback, usage results, and preferences. By using changes and system validation logs, the recommendation protocols can be improved over time. The aim is to provide a continuous improvement mechanism that enables optimization. Thus, the system... Validation and control without becoming directly dependent on unverified user input. It can improve its performance in generating recommendations based on data that has been processed through various steps. 9 In an alternative application of the invention, the different processing tasks involved in the system are: query analysis, information access, risk assessment, evidence analysis, discrepancy detection, verification, and recommendation. by multiple artificial intelligence agents responsible for functions such as production This can be accomplished, and the agents in question interact with each other through a coordination layer. It can exchange data. Thus, the system can utilize multi-agent artificial intelligence architectures with 5 It is becoming possible to implement it. In another alternative application of the invention, long-term user profile information, historical usage data, behavioral preferences, verified feedback, and over time The changes that occur can be stored within a dynamic digital user profile, and this profile It can help in establishing the transaction context in the proposal generation processes. 10 The invention also examines which information sources, which levels of evidence, and which sources the proposed solutions are based on. source versions, which risk assessments, which rule sets, and which validations by recording that it was created based on the results, the proposal process is retrospectively recorded. This enables monitoring. Thus, the technical auditing of system outputs is possible. reassessment as needed and under the same or updated data conditions 15 It is becoming possible to recreate it. In conclusion, the invention integrates user profile, contextual data, query, and risk status. evaluating hybrid information access in terms of evidence rating, timeliness, and version. management, resource conflict analysis, candidate scoring, confidence score calculation, and multi-stage 20 that combine with verification mechanisms; untrue or not supported by evidence Content that can be identified; personalized recommendations based on dynamic user profiles. capable of generating; and optimizing over time based on verified feedback; keeping track of transaction and resource traces and using different artificial intelligence, database and information management systems. a modular and scalable computer-aided system that can be implemented with its infrastructures and It offers a method. Explanation of the Figures: Figure 1: Verified proposal based on the invention profile, risk status, and hybrid information access. The overall architecture of the computer-implemented system that constitutes the protocol and the 5 fundamental components that make up the system. It refers to a schematic view showing the data and process relationships between modules. Description of Reference Parts: 100) System: User queries, user profile and contextual information, risk status and by processing information obtained from different information sources, it creates personalized, verified, a computer-based system that creates traceable and reliable recommendation protocols expression 10 is doing. 110) User terminal: The terminal through which the user accesses the system, through voice or written query. the electronic access unit that it created and received the suggestion protocol generated by the system It expresses. 120) Voice-to-text input interface: A system that receives voice or written input from the user, speech 15 The subject is to convert inputs into a processable data format and to perform preprocessing, natural language processing, and asset processing. It refers to the interface that transfers to the inference module (130). 130) Preprocessing, natural language processing and entity inference module: Preprocesses user input grammatically, By analyzing user intent, key concepts, entities, and queries from a semantic and contextual perspective. It refers to the module that defines the scope and other structured information related to the query. 20 140) Profile and context management engine: User profile, past usage data, user by evaluating preferences, authorization information, query and contextual information together 11 user-specific transaction context, based on newly verified data a profile that can be updated and a dynamic profile for use in subsequent recommendation processes It refers to the module that manages the data. 150) Risk classification and status machine: User profile, transaction context, and query. by evaluating the content, determining the risk score and / or risk level, and the determined risk is 5 Information access, filtering, evaluation, and verification conditions to be applied according to the level It refers to the managing module. 160) Layered information center: A system that manages different information sources and data repositories together and enabling the retrieval of information related to a user query from different information retrieval layers. It refers to the central data structure. 10 161) Vector index: The index between pieces of information found in information sources and a user query. Data that enables access to information related to a query by evaluating semantic similarity. It describes its structure. 162) Evidence-interaction knowledge graph: Knowledge assets, knowledge sources, products, contents, components, representing evidence and the relationships between them within a graphical data structure and word 15 It refers to a data structure that enables the evaluation of relationships. 163) Product / ingredient inventory matrix: By products, ingredients, active components, usage information regarding their accuracy, timeliness, and relationships between products or content It refers to a data structure that stores structured information. 170) Hybrid access and candidate pooling: 20 from different information sources and information access layers By combining the obtained results, candidate data sets are created for the user query. It refers to the module. 12 180) Constraint, scoring and candidate pruning engine: Candidate information sets based on user profile, risk level, safety regulations, compliance requirements, level of evidence, source reliability, timeliness, Evaluating, scoring, ranking, and ranking based on inter-source consistency and other technical criteria. It refers to the module that eliminates unsuitable candidates. 190) Protocol planning and production engine: Selecting candidate information sets to be suitable. 5 in accordance with user profile, transaction context, risk level and system operating rules by evaluating the module that forms the structure and content of the personalized recommendation protocol It expresses. 200) Two-stage security validator: Validates candidate datasets before and after proposal generation. The final proposed protocol created after the proposal creation process includes security, suitability, and evidence review. 10 It checks for consistency, source consistency, and verifiability; and also verifies falsehoods. It refers to a module capable of detecting content and unsupported information. 210) Response formatting module: Proposal that has successfully passed the validation process. converts the protocol into a user-friendly output format and sends it to the user terminal (110) It refers to the transfer module. 15 220) Source trace, version and audit log module: Used in the creation of the proposal protocol. information sources used, source versions, currency information, levels of evidence, applied rules, risk assessments, verification results, and transaction records It refers to the module that stores the information. 230) Feedback and expert approval module: Receives user feedback and approves that feedback. verifying and checking the notifications for consistency, and experts when necessary. feedback that guides evaluation and verification to the user profile and subsequent steps. It refers to the module that enables its use in proposal processes. 13 240) Generative model or rule-based engine: Determined and verified by the system. Based on the information sets, the final proposal protocol will be presented in natural language or in a structured format. its content is comprised of a generative artificial intelligence model, a rule-based decision engine, or a combination of these. It refers to the hybrid engine used. Detailed Description of the Invention: 5 The invention incorporates user profile information, contextual data, risk status, and user query information. By evaluating data from different information sources together, we create a hybrid information access system. processing methods, information obtained in terms of level of evidence, source reliability, timeliness, a recommendation that evaluates consistency between sources and compatibility with the user profile. Performing multi-stage verification before and after creation, and verification 10 As a result, creating reliable, traceable, verifiable, and personalized recommendation protocols. It relates to a computer-applied system and method. The system subject to the invention (100) transmits audio or written information to the system via the user terminal (110). receive user inputs and process those inputs through the voice-text input interface (120). It transfers the preprocessing, natural language processing and entity extraction module (130). 15 Preprocessing, natural language processing and entity extraction module (130), user-generated By analyzing the query from a grammatical, semantic, and contextual perspective, the key to understanding the query can be identified. concepts, entities, user intent, query scope, and other structured information It determines the information determined by the module (130) in question. It is converted into data and transferred to the profile and context management engine (140). 20 Profile and context management engine (140) converts structured query data into user profile data. information, past usage data, preferences, authorization information, and the query that was created by considering contextual conditions, the user-specific transaction context 14 This creates a system where the same user query can be applied to different user profiles or different... This allows for different interpretations depending on the contextual conditions. User profile used by the profile and context management engine (140); belongs to the user In addition to basic profile information, it also collects past usage data, preferences, and verified returns. This may include notifications and changes that occur over time. In this context, 5 profile and context management engine (140), verified new data with existing profile data By comparing these parameters, users can update specific parameters of their profile. Only verified or system-approved users can update their user profiles. Data with established reliability can be used; unverified user inputs are not profiled. Direct modifications to the data can be prevented. Thus, the user's 10 The dynamic profile is updated based on verified data. An alternative approach involves long-term user usage, behavior, preferences, and Verified feedback data, dynamic profile and context management engine (140) This can be represented by a digital user profile or the Digital Twin approach. The structure is a data model that represents the user's past and present state together. 15 It can be used to create the transaction context in the evaluation of subsequent queries. It contributes. Transaction context risk classification generated by profile and context management engine (140) and is transferred to the state machine (150). Risk classification and state machine (150), user profile, transaction context and query content 20 by jointly evaluating the risk score and / or risk level to be applied by the system The risk score is determined by the nature of the query, the risk parameters in the user profile, and the query itself. identified assets, product or content characteristics, terms of use, and system This can be calculated by taking into account other defined risk parameters. The system operates dynamically based on the determined risk score and / or risk level. The quality of information sources is determined according to the risk level, and candidate information sets are identified. The filters to be applied, the required level of evidence, the verification intensity, the confidence score thresholds, and The conditions for referring a case for expert evaluation may be changed. Thus, the system applies the same processing intensity to queries with different risk levels. 5 instead, differentiated verification processes depending on the determined risk level. is able to accomplish this. After the risk assessment is completed, the system becomes a layered information center (160) It initiates the information access process. Layered information center (160); vector index (161), evidence-interaction information graph (162) and 10 using the product / content inventory matrix (163) together or on separate data infrastructures It retrieves data related to the user query from different information layers. Vector index (161) is the sequence between the pieces of information in the information sources and the user query. Identifying pieces of information related to the query by evaluating semantic similarity. It provides. 15 Evidence-interaction knowledge graph (162), information assets, information sources, products, contents, components and represents the relationships between the evidence. Through this information graph (162) the source or evidence to which the information is related and the relationships between different information entities It can be determined. The evidence-interaction information graph (162) also includes the same or related 20 from different information sources. It can be used to compare information. Thus, differences or discrepancies between sources can be identified. Data can be generated to determine whether or not there are any discrepancies. 16 Product / ingredient inventory matrix (163), products, ingredients, active ingredients, usage information, structured information regarding their currency and the relationships between products or content. It contains information. Product / content inventory matrix (163) also includes products currently used by the user. or can be used to compare existing content with new candidate products or content. These 5 This includes obtaining the same active ingredient repeatedly from different products, or the use of identical or similar ingredients. Recurrence and identified product or content interactions can be detected. Information obtained by the layered information center (160) hybrid access and candidate pool It is transferred to module (170). Hybrid access and candidate pool module (170), semantic 10 obtained from vector array (161) similarity results, relational results obtained from the evidence-interaction information graph (162) and structured information obtained from the product / content inventory matrix (163) together It creates candidate information sets by evaluating them. Thus, the system is not dependent on only a single source of information or a single method of accessing information. without exception, information obtained from different data sources is used for a common candidate evaluation 15 It includes the generated candidate information sets in the process. Constraints, scoring, and candidate pruning are applied. It is transferred to the engine (180). Constraint, scoring and candidate pruning engine (180), candidate information sets are predefined Mandatory rules, security criteria, eligibility requirements, user profile, risk level, and preference. It is evaluated according to its parameters. 20 Candidate information sets are evaluated based on: semantic fit with the query, fit with the user profile, risk fit, and evidence. level, source reliability, source currency, consistency between sources, active ingredient repetition, Scores can be assigned based on product or content interactions and other technical criteria. 17 Candidate information sets can be ranked according to these scores and a predetermined threshold. Candidates who do not meet the requirements can be eliminated. This allows for protocol planning and production. only candidate data sets that meet the specified technical criteria can be used in the engine (190). The transfer is ensured. Using the source and evidence information contained in the evidence-interaction information graph (162), candidate 5 Evidence levels and / or evidence scores can be assigned to sets of information. The level of evidence... the nature of the information source and predetermined categories of evidence in determining It is available for use. These categories of evidence include expert opinion, clinical or technical guidelines, and systematic reviews. Meta-analysis may include research results and similar sources of information. The level of evidence is 10. in the scoring of the candidate information set and the confidence score calculated for the recommendation It is available for use. The presence of contradictions between the same or related information obtained from different sources. In this case, a source contradiction analysis can be performed. Within the scope of this analysis, the sources The level of evidence, reliability, timeliness, and version information can be compared. 15 Higher level of evidence, higher reliability, or more up-to-date information from conflicting sources. The source with the available version can be prioritized; in cases of unresolved discrepancies, the relevant information candidate is selected. They can be eliminated or redirected to the re-verification process. The resources contain layered information including publication date, update date, version information, and validity status. 20 via the central (160) and source trace, version and audit log module (220) This allows for evaluation. Thus, outdated or obsolete information sources can be used more effectively. Prioritization can be done among current resources. 18 The protocol planning and production engine (190) allows users to select suitable candidate datasets. by evaluating the profile, transaction context, risk level, and system operating rules. It creates a personalized recommendation protocol. Generative model or rule-based engine (240), protocol planning and production engine (190) Based on the structure determined by and the verified data sets, the final 5 of the proposal protocol It can generate text or structured output. The proposed protocol is passed to the two-stage security validator (200). Two-stage security validator (200), in the first validation stage the proposal candidate data sets to be used in the creation and second validation phase It oversees the final proposed protocol that has been created. 10 In the first validation phase, candidate data sets are evaluated based on suitability to the user profile, risk level, security rules, level of evidence, source currency, inter-source consistency, active ingredient It is checked in terms of repetition, product or content interactions, and other technical criteria. The product, content, components, and usage in the final recommendation generated during the second validation phase. Conditions, timing, quantity, frequency, and similar factors are being re-evaluated and verified. 15 They are compared with information sets. Within the scope of two-stage security verifier (200), also generative model or rule-based consistency of evidence between the content generated by the engine (240) and the supporting information sources It is being checked. Information that is not found in verified sources, and is not supported by sources... Contradictory statements or statements not supported by sufficient evidence can be identified, and such content will be removed. 20 deemed to be inaccurate content and subjected to a re-verification or reproduction process. It can be transferred. 19 As a result of this verification, the entire proposed protocol found to be flawed can be recreated. For example, by identifying the section of the suggestion where the error was detected, only that section is recreated. This can also be provided. Dynamic confidence score for each proposal and / or proposal component created within the scope of the invention. It can be calculated. The confidence score is based on: candidate score, level of evidence, source reliability, source 5. currency, consistency between sources, user profile compatibility, risk validation result, active parameters such as component and interaction controls and falsified content control results It can be calculated by considering them together. If the confidence score falls below the defined threshold, the relevant proposal will be re-evaluated. It is possible to obtain, sources with a higher level of evidence may be preferred, recommendation component 10 can be recreated or subjected to expert evaluation if necessary. It can be directed. The proposal protocol that passed the validation process is sent to the response formatting module (210) is being transmitted. The response formatting module (210) transmits the validated suggestion protocol to the user. It converts the output into a suitable output format that can be presented to the user via the terminal (110). 15 In alternative approaches, confidence score, level of evidence, source information, and verification status are also considered. It can be presented to the user along with the suggestion. Sources used in the creation of the proposal protocol, source versions, and currency. information, levels of evidence, applied rule sets, risk assessments, candidate scores, confidence Scores and verification results are recorded by the source trace, version and audit log module (220) 20 It is being recorded. Thus, each recommendation made should be based on specific sources, versions, and levels of evidence. and on what verification results it was based retrospectively It can be monitored. User feedback or items determined to require expert evaluation. The situations are transferred to the feedback and expert approval module (230). Feedback and expert approval module (230) directly incorporates user feedback into productive not transferring the feedback to the behavior of the model or rule-based engine (240); verifying the feedback, It undergoes consistency checks and, if necessary, expert evaluation. 5 Verified or expert-approved feedback, profile and context management. Updating the user profile and generating subsequent suggestions via the engine (140) It can be used in these processes. Verified feedback also includes previous recommendations and verification results. Candidate scoring, resource prioritization, profile fit, and similar systems are used for evaluation. 10 It can be used to optimize the parameters of the system over time. Thus, the system... Continuous improvement can be ensured through verified feedback. In an alternative application form, the system (100) includes query analysis, information access, risk different processes such as evaluation, evidence analysis, source contradiction analysis, verification, and recommendation generation. Multi-15 (multi-agent) where processing tasks are performed by different artificial intelligence agents Agent AI can be implemented with this architecture. In this case, the agents in question are part of the existing system. By exchanging data through its modules, it can share task results and final The recommendation can only be made based on results that meet the validation criteria. The modules used within the scope of the invention can run on a single server or in a distributed system. architecture, cloud computing infrastructure, or modular software running on different processing units 20 It can also be implemented in the form of structures. 21 Layered information center (160), vector index (161), evidence-interaction information graph (162) and product / content inventory matrix (163), different database technologies, information management systems or This can be achieved using data storage infrastructures. Generative model or rule-based engine (240), different major language models, artificial intelligence models, rule-based decision-making mechanisms or hybrid structures using both. 5 It can be accomplished. The invention involves a specific artificial intelligence model, database technology, programming language, or hardware platform. or not limited to communication infrastructure, user profile and risk assessment, hybrid information access, candidate knowledge assessment, evidence grading, source contradiction analysis, timeliness and version control, trust score calculation, false content detection, two-factor authentication (TFIA) Verification, dynamic profile updates, and continuous feedback based on verified data. It also includes equivalent technical applications that provide improvement functions.
Claims
22 REQUESTS 1- The invention incorporates user profile information, contextual data, risk status, and user data. Personalized and verified recommendation protocol by jointly evaluating the inquiry. It is a computer-based system that creates; its characteristic feature is: - at least one user terminal capable of receiving voice or text input from the user. 5 (110), - a voice-text input interface that receives inputs transmitted by the user (120), - a preprocessor that creates structured query data by analyzing user input, Natural language processing and entity extraction module (130), - Structured query data along with user profile and contextual information 10 a profile and context management engine that creates transaction context by evaluating (140), - risk score and / or risk level depending on the transaction context and query content a risk classification and status machine that determines (150), - at least one vector index (161), at least one evidence-interaction information graph (162) and at least one a layered information center (160) containing the product / content inventory matrix (163), 15 - candidate information by jointly evaluating the data obtained from the layered information center (160) a hybrid access and candidate pool module that forms clusters (170), - candidate data sets subject to predefined constraints, security rules, and compliance. evaluating, scoring, ranking, and deeming unsuitable according to the criteria a constraint that eliminates candidates, scoring and candidate pruning engine (180), 20 - a protocol that creates a proposal protocol from suitable candidate data sets planning and production engine (190), - candidate information sets prior to proposal formulation and the generated proposal protocol. a two-stage security validator that verifies after creation (200), 23 - a response that presents the verified recommendation protocol to the user in a suitable format. Formatting module (210), - a database that stores the information sources used, source versions, and validation records. Source trace, version and audit log module (220), - Verifying user feedback and subjecting it to expert review when necessary. a guiding feedback and expert approval module (230) and - the final text of the proposal protocol based on verified data sets or a generative model or rule-based engine that generates its structured output (240) It is characterized by its inclusion. 2- The invention is a computer-implemented system according to Claim 1, characterized by its profile and context. 10 the management engine (140), user profile information, past usage data, user by considering preferences, authorization information, and current contextual conditions together. It is characterized by creating a query-specific processing context. 3- The invention is a computer-implemented system according to Claim 2, characterized by its profile and context. management engine (140), verified new usage data, verified user 15 by evaluating preferences and verified feedback together with existing profile data It is characterized by the dynamic updating of the user profile over time. is being done. 4- The invention, according to Claim 1, is a computer-based system characterized by risk classification. and the state machine (150), depending on the query content, user profile and transaction context, 20 a risk score calculation and the assignment of that risk score to predetermined threshold values It is characterized by determining the risk level by comparison. 5- The invention, according to Claim 4, is a computer-based system characterized by risk classification. and the information access criteria of the state machine (150), depending on the determined risk level, 24 filtering conditions, required level of evidence, verification intensity, and confidence score threshold. It is characterized by its ability to determine its value dynamically. 6- The invention, according to Claim 1, is a computer-implemented system characterized by its proof-interaction feature. The information graph (162) includes information assets, information sources, products, contents, components, and evidence. It represents the relationships between information assets and the relevant information source and / or evidence. 5 It is characterized by its ability to associate information with other sources. 7- The invention, according to Claim 6, is a computer-implemented system characterized by constraints and scoring. and the candidate pruning engine (180), candidate information sets, the quality of information sources, Evaluation based on reliability and level of evidence, and the determined level of evidence for the candidate It is characterized by its use in calculating the score and / or confidence score. 10 8- The invention, according to Claim 1, is a computer-implemented system characterized by: evidence-interaction. information graph (162), constraint, scoring and candidate pruning engine (180) and resource trace, version and through the audit log module (220) the same or obtained from different information sources Identifying differences and / or inconsistencies between related information and providing evidence from sources. 15 characterized by being prioritized according to level, reliability, up-to-dateness and version information. is being done. 9- The invention, according to Claim 1, is a computer-implemented system characterized by layered information. central (160) and source trace, version and audit log module (220), information sources By evaluating the release date, update date, version information, and validity status, it is up-to-date. and valid sources compared to outdated and / or obsolete sources 20 It is characterized by its prioritization. 10- The invention, according to Claim 1, is a computer-implemented system, characterized by its product / content. inventory matrix (163), products and / or currently used by the user By comparing the contents with the candidate product(s) and / or contents, the same active ingredient can be found in different products. repeat purchase and predetermined interactions between products and / or ingredients It is characterized by its ability to detect. 11- The invention, according to Claim 1, is a computer-implemented system characterized by constraints, scoring, and candidate pruning engine (180), candidate information sets semantic suitability, user profile compatibility, risk compatibility, level of evidence, source reliability, source currency, inter-source compatibility. scoring based on at least two of the following criteria: consistency and product / content compatibility. by ranking the candidates and eliminating those that do not meet a predetermined threshold value It is characterized by... 12- The invention, according to Claim 11, is a computer-implemented system, characterized by constraints and scoring. and the candidate pruning engine (180) requires 10 for each candidate information set and / or candidate information component. Calculating a dynamic confidence score using candidate score and reliability parameters. It is characterized by... 13- The invention, according to Claim 12, is a computer-implemented system characterized by dynamic reliability. candidate score, level of evidence, source reliability, source currency, cross-referenced sources Consistency, user profile compatibility, risk validation results, and two-factor authentication 15 at least two of the verification results performed by the validator (200) together It is characterized by its evaluation and calculation. 14- The invention, according to Claim 1, is a computer-implemented system characterized by its two-stage operation. security validator (200), candidate information sets in the first validation phase User profile suitability, risk level, security rules, level of evidence, source 20 in terms of timeliness, consistency between sources and product / content compatibility, and secondly the final proposed protocol created during the validation phase, verified data sets and It is characterized by its verification process by comparing it with system rules. 15- The invention, according to Claim 14, is a computer-implemented system characterized by its two-stage operation. 25 by the security validator (200), generative model or rule-based engine (240) 26 the final content created and the verified information sources on which that content is based to check the consistency of evidence between them and its counterpart in verified information sources Identifying content that is missing, contradicts sources, or is not supported by sufficient evidence. It is characterized by... 16- The invention, according to Claim 15, is a computer-implemented system characterized by a two-stage 5-phase system. Content that cannot be verified by the security verifier (200), evidence inconsistency, security a discrepancy with the criteria or a result below the determined confidence score threshold was detected. If this happens, the entire proposed protocol or the part where the non-conformity is found will be rejected. by sending the protocol back to the planning and production engine (190) and recreating it It is characterized. 10 17- The invention, according to Claim 1, is a computer-implemented system, characterized by its source trace and version. and the information resources of the audit log module (220), related to the proposed protocol created, source versions, up-to-date information, evidence levels, risk assessments, candidate recording scores, confidence scores, and verification results by correlating them with each other. It is characterized by its ability to take control. 15 18- The invention, according to Claim 1, is a computer-implemented system characterized by feedback and The expert approval module (230) uses user feedback in the generative model. or verification and consistency check before being transferred to the rule-based engine (240) by having the patient undergo the procedure and referring them to an expert for evaluation if deemed necessary. It is characterized by... 20 19- The invention, according to Claim 18, is a computer-implemented system, characterized by its verified nature. and / or expert-approved feedback is entered into the profile and context management engine. (140) transferring and updating the user profile with the said feedback, in improving candidate scoring criteria and subsequent proposed protocols It is characterized by its use in its creation. 25 27 20- The invention, according to Claim 1, is a computer-implemented system whose feature is query analysis, information access, risk assessment, evidence analysis, source contradiction analysis, candidate information evaluation, verification, and recommendation generation processes, at least two of which are performed by different artificial intelligences. the performance by agents and the existing system of these artificial intelligence agents Multi-agent (Multi-5) is characterized by sharing processing results through its modules. It has an Agent AI architecture. 21- The invention is a computer-implemented system according to Claim 2 or 3, characterized by its profile and context management engine (140) long-term usage, behavior, preference of the user and by representing verified feedback data within a dynamic user profile 10 characterized by its use of the subject profile in evaluating subsequent queries. It has a digital user profile (Digital Twin) structure. 22- The invention is a method implemented by a computer, and its characteristic is; - receiving a voice or written query from the user, - the received query is structured through preprocessing, natural language processing, and entity extraction. conversion to query data, 15 - structured query data along with user profile and contextual information. by evaluating and establishing the context of the transaction, - risk score and / or risk level depending on the transaction context and query content determination, - Determining the conditions for accessing and verifying information based on the identified risk level, 20 - from vector index (161), evidence-interaction information graph (162) and product / content inventory matrix (163) access to information is realized, - Candidate information obtained through hybrid access and candidate pool module (170) converting into clusters, 28 - candidate information sets constraint, scoring and candidate pruning engine (180) evaluation, scoring, ranking, and elimination of unsuitable candidates, - Development of a proposal protocol from suitable candidate information sets, - Recommendation generation by the two-stage security validator of the recommendation protocol (200) Verification before and after, 5 - Checking the consistency of the proposed content with supporting information sources and evidence. - detection of unverified and / or false content, - calculating the confidence score for the recommendation and - steps to present the user with a proposed protocol that meets the verification criteria It is characterized by its inclusion. 10 23- The invention, according to Claim 22, is a method implemented by a computer, characterized by: Identifying discrepancies between identical or related information obtained from different sources. the level of evidence, reliability, currency, and version information of the sources are evaluated. comparison and prioritization of candidate knowledge sets based on identified resource priorities. It is characterized by its use in evaluation. 15 24- The invention is a method implemented by a computer, according to Claim 22 or 23, This feature incorporates user-provided feedback and provides expert approval. Verification by module (230) and / or expert evaluation and Verified feedback will be used to update the user profile and then make recommendations. It is characterized by its use during the creation of protocols. 20 25- The invention involves storing data on a computer-readable medium and one or more When executed by multiple processors, the method defined in Claims 22 through 24 computer program instructions that enable the execution of a task It is a software product, and its characteristic feature is that it allows the user to follow the instructions of the computer program in question. processing of the query, evaluation of user profile and risk level, hybrid 25 29 facilitating information access, scoring and eliminating candidate information sets, calculation of the confidence score, two-stage verification of the proposal protocol, evidence ensuring consistency and verifying false information, and providing verified recommendations. Its purpose is to ensure the creation of the protocol.