SMART PROCUREMENT AND CONTRACT MONITORING SYSTEM

Abstract

It is related to a system (1) that enables matching of order and contract items using natural language processing methods, detection of price deviations using statistical and unsupervised learning techniques, risk analysis taking into account supplier performance, suggestion of alternative products and preparation of reports for managers on the subject with artificial intelligence-based scoring.

Description

1 TARIFF SMART PROCUREMENT AND CONTRACT MONITORING SYSTEM Technical Area This invention enables the processing of order and contract items using natural language processing methods. matching, price deviations using statistical and unsupervised learning techniques to identify and conduct risk analysis taking into account supplier performance, Suggesting alternative products and providing managers with AI-based scoring. It relates to a system that enables the preparation of reports on the subject. 10 Previous Technique Existing procurement management and contract compliance control systems, procurement their operations are mostly carried out using rule-based methods and static matching mechanisms. 15 It evaluates using these systems, order items and contracts. The relationship between the items is usually based on keywords, product codes, and category information. or is established through predefined catalog records, and is technical and Semantic similarities cannot be analyzed effectively. The systems used... They mostly focus only on price comparison and do not examine the technical aspects of the products. characteristics, supplier performance, past purchasing behavior, and the contract. The eligibility criteria cannot be evaluated holistically. This situation affects the purchase. efficiency losses in procurement processes, delayed identification of supplier-related risks This leads to missed opportunities for cost optimization. Therefore, 25 next-generation decision support systems will be used in purchasing processes. It is needed. United States patent number US20210201013A1 in the known state of the art in the document, the management of contract data in digital environment, contract analysis of the contents, monitoring of the contract lifecycle and contract 30 2 a system developed for the purpose of executing workflows related to processes It is explained. Brief Description of the Invention The purpose of this invention is to combine item data and contract data related to purchasing transactions. Performing a contract compliance check by semantically matching prices. and identifies supplier-related deviations, conducts risk assessments, and makes decisions. The goal is to create an AI-based system that generates support suggestions. Another aim of this invention is to reduce potential cost increases in purchasing processes, by identifying contract non-conformities and operational risks at an early stage to offer explainable suggestions to the user and alternative technically compatible products by providing suggestions, the accuracy and efficiency of decision-making processes to increase. 15 Detailed Description of the Invention The "Smart Procurement and Contracting" project was carried out to achieve the purpose of this invention. The "Audit System" is shown in the attached figure; Figure 1. Schematic view of the system that is the subject of the invention. 20 The parts shown in the figure are individually numbered, and the corresponding numbers correspond to these numbers. given below: 1. System 25 2. Electronic device 3. Database 4. Server 3 Semantic, statistical, and artificial intelligence analysis of procurement and contract compliance. evaluation using intelligence-based methods, risk analysis, and decision-making. The system in question, which enables the production of supporting outputs (1); - for entering order requests and purchasing processes It provides an interface that allows viewing and communication with remote servers. 5 at least one electronic device capable of setting up (2), - operational, analytical and learning data, applied to orders and order processes data regarding, contract data, supplier information and user at least one data point structured to enable the storage of actions base (3) and 10 - purchase entered by the user via electronic device (2) receiving requests, processing the texts related to the order items and interpreting them semantically. Creating vector representations, semantic analysis of contract items matching, contract data with anomaly detection algorithms the analysis, processing of supplier performance data and risk scores 15 the creation of a multidimensional risk assessment by combining all the obtained scores. an evaluation should be made, alternatives based on technical specifications data. generating product recommendations, evaluating the economic suitability of the products based on these recommendations evaluation using predictive models, and user feedback. updates should be made, all results should be explained by AI 20 the methods should be used to justify the presentation and all outputs should be data. It includes at least one server (4) that enables recording to the database (3). The electronic device (2) in the system (1) which is the subject of the invention, has an ERP system 25 of the purchase requests created by internal users through the system at least one interface that allows access and viewing of purchasing processes It is structured to present the electronic device (2), known technology. using any remote communication protocol included in the case of the server (4) and establish a connection with the database (3) and through this connection to the server (4) 4 It is structured to enable data exchange. The invention in its preferred application electronic device (2) server (4) and data bus It is configured to exchange data using the internet. The database (3) in the system (1) which is the subject of the invention is in communication with the server (4) 5 and is configured to be managed by the server (4). The invention is preferred In the arrangement made, the database (3), purchases entered through the ERP system Product name, dimensions, power, voltage, etc., all technical specifications related to the requests, product unit price and quantity data, item names of active official contracts within the company, product prices and expiration dates, numerical data vectors, FAISS (Facebook 10 calculated via an artificial intelligence similarity search library (AI SIMPLE) search engine cosine similarity matrices, generated semantic similarity scores, ordering historical price movements of the products, Z-score (standard score), MAD calculated using median absolute deviation and IQR (interquartile range) methods abnormal price deviation rates, past delivery times of product suppliers, product 15 performance scores, Isolation Forest, One-Class SVM (one-class (Class-based Support Vector Machine) and DBSCAN (Density-based Spatial Clustering) anomaly detection scores generated with (clustering with noise applications) models with classification labels such as "suitable", "should be monitored" or "high risk", Feature 20 that is given meaning through organizational ontologies in risky situations vectors, alternative equivalent product information, cost advantage verification records, SHAP values, presented decision support reports, screened by the expert. Transaction decisions in the form of approval or rejection are made via ETL (data extraction, (transformation and upload) flowcharts and IT (information technology) infrastructure data It is configured for storage. 25 The server (4) in the system (1) which is the subject of the invention, any remote communication to communicate with the electronic device (2) using the protocol and the established protocol to exchange data with electronic devices (2) via communication It is being configured. The server (4) records new data into the database (3), Deletion of registered data in database (3) or database (3) 30 Modifying the registered data within and the registered data within the database (3) to manage the database (3) by operations such as updating the data It is configured. The server (4) receives the product order via the electronic device (2). It is configured to receive the request. The server (4), user Product order 5 entered via ERP (enterprise resource planning) It is configured to receive the data. The server (4) logs in via ERP. current order information including product name, specifications and price It is configured to enable the receipt of the user's input. The server (4) The server (4) is configured to process the order data of the suppliers. It is configured to verify the current order data entered. Server (4), 10 to compare newly created order requests with active order processes The server (4) is configuring the order data to compare with the processed order data. item names, contract prices, and validity period of active contracts. The server (4) is configured to retrieve the information from the database (3). Of the contract data taken from the base (3), 15 are still valid. to identify contracts and only retain valid contract records It is structured to include in the comparison process. The server (4) is made semantic differences between order items and contract items through comparison to identify mismatches, detect non-compliance with the contract and abnormal deviations It is structured to perform and produce decision support outputs. Server (4), 20 the product name and specifications entered by the supplier in free text format, by cleaning with natural language processing (NLP) algorithms and converting them into numerical vectors It is configured to convert the resulting order vectors. The server (4) Quick similarity search of contract item vectors retrieved from the database (3) It is configured to compare with the engine (FAISS). Server (4), 25 to calculate the cosine similarity metric as a result of the comparison It is structured. Server (4), cosyn similarity and semantic similarity score It is configured to produce. Server (4) via cosine similarity. By analyzing the image, the supplier can find the image that is semantically most similar to the product provided. It is structured to identify the contract item. Server (4), semantic 30 as per the matching item, with the current price requested by the supplier and the official price. 6 It is configured to compare contract prices. The server (4) to analyze the past price movements of the product in question is being configured. The server (4) displays the history of the product item for which an order request was created. The server (4) is configured to retrieve price data from the database (3). Z- Detecting abnormal price deviations using score, MAD and IQR statistical methods 5 The server (4) is configured to perform Z-score, MAD and IQR statistical analysis. It is structured to calculate abnormal price deviations using various methods. Server (4), past delivery time of the product suppliers from which purchases were made, to obtain performance scores based on material quality and contract compliance history It is structured accordingly. The server (4), product suppliers' delivery time, 10 performance scores based on material quality and contract compliance history It is configured to normalize the calculated semantics. The server (4) similarity score, price variance ratio and supplier performance score outputs by evaluating with Isolation Forest, One-Class SVM, and DBSCAN algorithms. It is configured to detect anomalies. Server (4), calculated 15 semantic similarity score, price variance ratio and supplier performance score together to create an anomaly detection result by evaluating and the anomaly result obtained It is structured to trigger an alternative product suggestion process accordingly. Server (4) marks the purchase transactions that have undergone anomaly detection analysis as “suitable”. 20 to classify and label as "should be monitored" or "high risk" is configured. The server (4) considers the purchase process to be risky or expensive. In the case of classification, the dimensions, power, and voltage of the product entered by the supplier are specified. to make sense of its characteristics in this way by using institutional ontologies is structured. The server (4) interprets these feature vectors into KNN (K- By using the Nearest Neighbors (K-nearest neighbors) algorithm, 25 to choose a cheaper alternative product that will perform a similar and identical function It is being structured. The server (4) will discuss the future of alternative equivalent products to be proposed. price predictions are made using a random forest-based regression algorithm (Random The server (4) is configured to calculate via Forest Regressor. 30 to verify that the alternative product selected is cost-effective. is configured. The server (4) displays the obtained anomaly classification labels, 7 calculated price variance rates and cost advantage data of the selected equivalent product It is configured to bring together the Server (4), XGBoost (eXtreme Purchase with Gradient Boosting (extreme gradient boosting algorithm) The server (4) is configured to calculate the final risk score. The data is configured to be used as an XGBoost input parameter. 5 The server (4) uses the generated final risk score to optimize purchasing processes for economic products. situations that it deems risky by examining them in terms of quality and equivalent products. It is structured to rate. Server (4), the final risk obtained. SHAP explains the score and the mathematical reasons behind the obtained score. (Shapley Additive exPlanations) values ​​and 10 a decision support report by combining the proposed economically viable alternative products. It is configured to create. The server (4) creates decision support. and in the report, to explain the parameters that affect the risk score. The server is configured to present the reasons for the recommendation to the user. (4), decision support reports and risk alerts, Power BI (business intelligence and data visualization 15 business intelligence platforms such as Grafana (data visualization and monitoring platform) to present to the purchasing specialist via the panels and obtain approval It is configured. The server (4) displays the risk and purchasing specialist's screen. After reviewing alternative proposals, the supplier approves or rejects the input. The server (4) is configured to make the transaction decision in this way. the time the expert spends reviewing the recommended products, their explanations regarding the recommendation, and It is configured to record the transaction reasons. Server (4), The generated recommendation, recommendation outcome, user feedback, and recommendation success rate data. to store the recommendation history dataset in the database (3) It is being configured. Server (4), 25 for equivalent product by the purchasing specialist. The server (4) is structured to receive the results of the given decision. Q-learning (Q-) marks decision outcomes as reward or punishment signals. Learning (Q-learning) and DQN (Deep Q-Network) algorithms It is configured to feed the feedback received from users. The server (4) Optimize the AI ​​recommendation policy over time based on feedback. 30 It is configured to do so. The server (4) is more suitable for subsequent purchase processes. 8 It is structured to provide accurate product recommendations. Server (4), all data entries, analysis results, risk reports, and expert opinions that have been processed. their decisions are saved to the database for reuse in subsequent processes (3) It is configured to save. The server (4) returns the accumulated user data. AI 5 uses notifications, recommendation history data, and transaction results. The models are configured to retrain at specific intervals. Industrial Application of the Invention The system in question (1) is designed to address deviations in instantaneous data streams and In-depth analysis of the technical impacts of operational decisions on infrastructure 10 by creating a closed-loop feedback loop, thus enabling itself dynamically integrates AI-based decision algorithms within its structure. updating and enabling real-time monitoring of changing corporate dynamics. This is made possible by the invention's architecture, which enables processing-intensive applications in logistics, finance, and related fields. only risk or irregularities in production and enterprise resource planning infrastructures 15 No detection is being made, and at the same time, the flowcharts in the data warehouse layer, the data Prioritization processes in processing queues are being restructured. This invention ensures the continuity of service and Fully compatible with different industrial systems, horizontally scalable and operational. From this perspective, it provides a sustainable system (1) optimization. 20 Based on these fundamental concepts, the invention is called "Smart Procurement and Contract Control". It is possible to develop a wide variety of applications related to the System (1)”, and the invention This cannot be limited to the examples described here, but is primarily stated in the claims. It is like this. 25

Claims

9 REQUESTS 1. Semantic, statistical and analytical analysis of procurement transaction and contract compliance. evaluation and risk analysis using artificial intelligence-based methods and enables the production of decision support outputs; 5 - for entering order requests and purchasing processes It provides an interface that allows viewing and communication with remote servers. at least one electronic device capable of setting up (2), - operational, analytical and learning data, applied to orders and order processes data regarding, contract data, supplier information and user 10 at least one data point structured to enable the storage of actions base (3) and - purchase entered by the user via electronic device (2) receiving requests, processing the texts related to the order items and interpreting them semantically. Creating vector representations, contract items and semantics 15 matching, contract data with anomaly detection algorithms the analysis, processing of supplier performance data and risk scores the creation of a multidimensional risk assessment by combining all the obtained scores. an evaluation should be made, alternatives based on technical specifications data. generating product recommendations, evaluating the economic suitability of the products based on these recommendations 20 evaluation using predictive models, and user feedback. updates should be made, and all results should be explained using artificial intelligence. the methods should be used to justify the presentation and all outputs should be data. with the inclusion of at least one server (4) that enables recording to the database (3). a characterized system (1). 25 2. Created by internal company users via the ERP system. entering purchase requests and purchasing processes configured to provide at least one interface that allows its display A system like the one in Claim 1, characterized by an electronic device (2) (1). 30 3. Using any remote communication protocol, the server (4) and the database (3) to establish a connection and send data to the server (4) through this connection. electronic device configured to enable the purchase to take place (2) A system like those in Claims 1 and 2, characterized by (1).

4. Data exchange with the server (4) using the Internet as a data bus. characterized by an electronic device (2) configured to perform a system like the one in Request 3 (1).

5. Communicating with server (4) and being managed by server (4) 10 from the above requests characterized by the structured database (3) a system like any other (1).

6. Product name, dimensions, power, for purchase requests entered through the ERP system, All technical specifications such as voltage, product unit price and quantity data, company 15 item names and product prices of active official contracts within its organization. validity periods, numerical data vectors, FAISS search engine cosine similarity matrices calculated from the generated semantics similarity scores, historical prices of ordered products abnormal price movements are calculated using Z-score, MAD, and IQR methods. deviation rates, past delivery times of product suppliers, product performance scores, Isolation Forest, One-Class SVM (single-class support anomaly detection scores generated with vector machine and DBSCAN models with classification as "suitable", "should be monitored" or "high risk" labels, in risky situations, through corporate ontologies 25 interpreted feature vectors, alternative equivalent product information, financial advantage validation records, SHAP values, decision support provided reports are approved or rejected by the expert via the screen. recording transaction decisions in the form of ETL flowcharts and infrastructure data. database structured to ensure that it is kept under (3) and 30 11 a system like any of the above characterized claims (1).

7. Receiving the product order request entered via the electronic device (2). The above 5 is characterized by the server (4) configured to provide a system like any of the requests (1).

8. Entered by the user via the ERP (enterprise resource planning) system. characterized by the server (4) configured to receive product order data. a system like any of the above requests (1). 10 9. Current information including product name, specifications, and price entered via the ERP system. server configured to enable order information to be received (4) a system like any of the above characterized claims (1). 15 10. Server configured to process the order data entered by the user (4) as in any of the above claims characterized by system (1).

11. To verify the current order data entered by suppliers 20 from the above requests characterized by the configured server (4) a system like any other (1).

12. Compare newly created order requests with active order processes. The above 25 is characterized by the server (4) configured to do so. a system like any of the requests (1).

13. Active contracts for comparison with processed order data. data including item names, contract prices, and validity period information. 30 contract data taken from the base (3) and from the database (3) to identify contracts that are still in effect and only those that are valid 12 to include ongoing contract records in the comparison process from the above requests characterized by the configured server (4) a system like any other (1).

14. The comparison revealed a 5% difference between order items and contract items. Identifying semantic matches, contractual inconsistencies, and anomalies. structured to detect deviations and produce decision support outputs any of the above requests characterized by the server (4) such a system (1).

15. The product name entered by the supplier in free text format, and by cleaning its features using natural language processing (NLP) algorithms, numerical characterized by the server (4) configured to convert to vectors. a system like any of the above-mentioned requests (1).

16. The order vectors obtained and the contract items taken from the database (3) to compare vectors using the Fast Similarity Search Engine (FAISS) and Calculating the cosine similarity metric based on the comparison results. The above is characterized by the server (4) configured to do so. a system like any of the requests (1). 20 17. Generating a semantic similarity score using cosine similarity and cosine By analyzing similarities, the supplier's input into the product is semantically analyzed. to identify the most similar formal contract item. 25 of the above requests characterized by the configured server (4) a system like any other (1).

18. Based on the semantically matching item, the current information requested by the supplier. to compare the price with the official contract price and the product item in question Server 30 is configured to analyze past price movements. 13 (4) like any of the above-mentioned claims characterized by system (1).

19. Data on the historical price of the product item for which an order request has been created. from base (3) and using Z-score, MAD and IQR statistical methods 5 to detect abnormal price deviations and calculate Z-score, MAD and IQR to calculate abnormal price deviations using statistical methods from the above requests characterized by the configured server (4) a system like any other (1).

20. Past delivery times and materials of product suppliers from whom purchases were made. to receive performance scores based on quality and contract compliance history The above is characterized by the server (4) configured to do so. a system like any of the requests (1).

21. Product suppliers' delivery time, material quality and contract compliance to normalize performance scores based on past performance from the above requests characterized by the configured server (4) a system like any other (1).

22. Calculated semantic similarity score, price variance ratio and supplier By evaluating performance score outputs, Isolation Forest, One-Class SVM is used for anomaly detection with DBSCAN algorithms. from the above requests characterized by the configured server (4) a system like any of them (1). 25 23. Calculated semantic similarity score, price variance ratio and supplier anomaly detection result by evaluating performance scores together to create and alternative products based on the anomaly results obtained. 30 characterized by the server (4) configured to trigger the proposal process a system like any of the above-mentioned requests (1). 14 24. Purchases that have undergone anomaly detection analysis are categorized as "appropriate" or "should be monitored". or to classify and label as "high risk" from the above requests characterized by the configured server (4) a system like any of them (1). 5 25. If the purchase transaction is classified as risky or expensive, The supplier's corporate specifications for the product, such as dimensions, power, and voltage, must be met. The server (4) is structured to make sense using ontologies. A system like any of the above characterized claims 10 (1).

26. These feature vectors, which have been interpreted, are called KNN (K-Nearest Neighbors – K-en by processing it using the nearest neighbors algorithm, it finds similar and identical objects. They will see the cheaper equivalent product and choose the alternative that will be suggested. Future price predictions for equivalent products are based on a random forest-based 15 server configured to perform calculations via regression algorithm (4) like any of the above-mentioned claims characterized by system (1).

27. To verify that the selected alternative product is financially advantageous and 20 the anomaly classification labels obtained, the calculated price deviation combining the ratios and cost advantage data of the selected equivalent product The above is characterized by the server (4) configured to do so. a system like any of the requests (1).

28. Calculate and obtain the final risk score for the purchase transaction with XGBoost. to use the obtained data as XGBoost input parameters from the above requests characterized by the configured server (4) a system like any other (1). 15 29. The final risk score created will be used to improve purchasing processes and product quality. and by examining it in terms of equivalent products, it finds situations that are risky. characterized by the server (4) configured for rating a system like any of the above requests (1).

30. The final risk score obtained and the factors behind that score. Finding the SHAP values ​​that explain the mathematical reasons, A decision support report combining economically viable alternative product recommendations. to create and influence the risk score in the generated decision support report. 10 characterized by the server (4) configured to present to the user a system like any of the above requests (1).

31. Decision support reports and risk alerts can be accessed through Power BI and Grafana. Presenting and obtaining approval from the purchasing specialist via business intelligence panels 15 The above is characterized by the server (4) configured to do so. a system like any of the requests (1).

32. The purchasing specialist reviews the risks and alternative alternative recommendations on the screen. 20 and the time it takes for the purchasing specialist to review the recommended products, according to the recommendation. to record their statements and the reasons for the transaction from the above requests characterized by the configured server (4) a system like any other (1).

33. Suggestions generated, suggestion results, user feedback, and suggestion success rate. to store the data in the database as a recommendation history dataset (3) from the above requests characterized by the configured server (4) a system like any other (1). 16 34. The results of the decision made by the purchasing specialist regarding the equivalent product. receiving and marking the results of the given decisions as a reward or punishment signal. Server configured to feed Q-learning and DQN algorithms (4) as in any of the above claims characterized by system (1). 5 35. AI recommendations based on user feedback. to optimize its policy over time and for subsequent purchasing processes with the server (4) configured to provide more accurate product recommendations A system like any of the above characterized claims 10 (1).

36. All data entries, analysis results, risk reports, and expert decisions can be reused as data in subsequent processes. characterized by the server (4) configured to save to the base (3). a system like any of the above requests (1). 15 37. Accumulated user feedback, suggestion history data, and transaction results. using artificial intelligence models to retrain them at specific intervals from the above requests characterized by the configured server (4) a system like any other (1). 20 30

Images