ANOMALY DETECTION SYSTEM
Patent Information
- Application Number
- TR202613580
- Authority / Receiving Office
- TR · TR
- Patent Type
- Applications
- Current Assignee / Owner
- Filing Date
- 2026-08-11
- Publication Date
- 2026-09-21
Smart Images
Figure 00000014_0000
Abstract
Description
1 TARIFF ANOMALY DETECTION SYSTEM Technical Area This invention allows for 5 interactions with the same customer in call centers and customer service systems. Analysis of related sequential calls in terms of content, time, and behavior. This is done by making duplicate calls that deviate from normal customer feedback. with a system that enables the detection of anomalies based on the large language model (LLM). It is related. Previous Technique In the known state of the art, call centers and customer service systems Detection of duplicate customer calls and unusual call behavior Solutions focused primarily on call volume, call frequency, and average call duration. and is based on numerical statistics such as the time difference between calls. Current Using simple statistical methods in systems, the same customer within a specific time interval if the number of calls made by exceeds fixed threshold values A repeat call or unusual situation detection is being carried out. Other available information In solutions, call to action texts use keyword matching or predefined subject lines. 20 They are classified according to their headings and call densities for specific topics. This is being monitored. However, these approaches do not focus on the semantic content of the call for proposals and It does not take into account the contextual continuity between calls, only the words matching at the level or individual evaluation of calls It is based on. Therefore, a whole of 25 consecutive calls associated with the same customer This involves considering and analyzing recurring call patterns based on content. It is technically not possible. This situation applies to voice calls. acoustic characteristics such as speech rate, intonation, or mood that can be inferred The fact that it is not included in the anomaly detection process, and that the searches conducted are only quantitative... evaluation through metrics and qualitative deviations in call behavior 30 This makes it impossible to detect, and therefore it needs to be context-sensitive and automatic. 2 A new technical solution is needed that allows for the detection of anomalies in this way. It is heard. United States Regulation US10965807B2, regarding the known state of the art. Machine learning for detecting anomalies or fraudulent behavior in patent documents 5 content that can be used directly, also specific to a call and user relationship. a multimodal behavioral analysis scoring approach that produces a score equivalent to the information. We are talking about a system that provides this. Brief Description of the Invention 10 The aim of this invention is to eliminate duplicate customer calls in call centers. Customer rating is determined based on call volume and fixed thresholds. anomalies that cause disruptions in the service process due to feedback The aim is to provide a system that enables the differentiation of patterns. 15 Detailed Description of the Invention The "Anomaly Detection System" implemented to achieve the purpose of this invention is attached. as shown in the figure; 20 Figure 1. Schematic view of a system that is the subject of the invention. The parts shown in the figures are individually numbered, and these numbers... The corresponding answers are given below. 25 1. System 2. Electronic device 3. Database 4. Server 30 3 Consecutive customer calls are analyzed based on content, semantic similarity, and temporal continuity. by analyzing them together in terms of repetitive behavior that deviates from normal behavior The invention is a system that enables the automatic detection of calls (1); - information on call data from customers, call analysis Viewing the request and the results of the analyzed call data 5 at least one electronic device that provides an interface structured to (2), - contains customer call records, customer data, and temporal data. call sequences, contextual indicator data, anomalies, and anomaly at least one 10 configured to store classification parameters database (3) and - using any remote communication protocol electronic device (2) communicating with, responding to call anomaly analysis requests from the user. to obtain call and customer information received from call centers 15 to collect and synchronize this information over time, Pre-processing missing or erroneous records and information in the collected data. filtering pre-processed data by time and customer. creating call sequences by grouping, within the call sequence By analyzing and categorizing speech transcripts with LLM, every 20 Creating a vector representation for calls, from audio recordings of calls extracting acoustic and emotional characteristics, and interpreting these characteristics textually. Obtaining multimodal feature vectors by combining them with representations, Trust is established by classifying user intent from the content of incoming calls. Obtaining scores, tracking all calls, and ensuring solution continuity 25 and to assess behavioral deviations and detect anomalies and calculating the anomaly score, calling up the scores and other data obtained Monitoring and decision support by integrating into central management systems at least one structured to enable its use in processes The server includes (4). 30 4 In the system that is the subject of the invention (1) electronic device (2), smartphone, tablet computer, It is a device similar to a desktop computer or a portable computer. The one in question... electronic device (2) any remote device in the known state of the art to establish a connection with the server (4) using the communication protocol and with the server (4) It is configured to perform data communication. Electronic device (2), intermediate 5 Call logs, user information, and multimodal data entered via the face. transfer of components to the server (4) and LLM run on this data Viewing the anomaly detection outputs generated as a result of base-based analyses. It is structured to provide the preferred application of the invention. electronic device (2) server (4) using Internet as data bus 10 It is structured to facilitate the purchase. In the system that is the subject of the invention, the database (3) in (1) is in communication with the server (4) and is configured to be managed by the server (4). The invention is preferred In the arrangement made, the call learned through the database (3), server (4) 15 records, customer information, timestamps, speech transcripts, and audio data and the semantic outputs derived from them, acoustic properties, multimodal representations, time differences between calls, repetition intensity data, customer emotional state data, customer speech rate, energy and intonation information, between calls Similarity scores, resolution state continuity and acoustic-emotional indicators, 20 anomaly scores, caller ID numbers, call sequence relationship information, call array indexing data, data synchronization records, customer intent, problem status, main category and subcategory data for each call, parameters configurations, threshold values and storage of feedback records It is structured to provide. 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) adds new data to the database (3) 30 recording, deleting the registered data in the database (3) or data Modifying the registered data in the database (3), within the database (3) Data processing involves operations such as updating and retrieving recorded data. It is structured to manage the base (3). The preferred method of the invention In the arrangement of the server (4), call data coming through the electronic device (2) After receiving the request for analysis, call 5 is processed through call center infrastructures. records, timestamps, customer identification information, automated speech recognition obtaining speech transcripts and audio data produced through the systems and is configured to ensure that it is recorded in the database (3). Server (4) common by preprocessing the data recorded in the database (3). synchronize on the timeline, extract missing or erroneous records, and 10 It is configured to make it suitable for analysis. The server (4), before By grouping processed call data based on customer ID, it is sorted chronologically. to sequence, calculate time differences between consecutive calls and in advance Calls that fall within a defined sliding time window are the same call. It is configured to be included in the array. Server (4), time window 15 To start a new call sequence for calls outside of this sequence and assign each call to a specific method... within a temporal interaction set that arises around the customer problem It is structured to model. The server (4) is located within each call sequence. field speech transcripts big language model (LLM) based systems (1) By analyzing through, the main category, subcategory, customer intent, problem 20 to produce semantic outputs consisting of situational and contextual indicators, this To generate semantic vector representations for each call from the outputs, and these vectors The representations are configured to be saved to the database (3). The server (4) data semantic between vector representations and calls taken from base (3) It is structured to enable comparisons to be made. 25 Server (4) receives data from electronic devices (2) in addition to textual analyses. Voice data from calls includes speech rate, voice energy, intonation, and emotion. to extract the acoustic characteristics of the situation and to translate these characteristics into textual representations by combining them to create multimodal feature vectors for each call. is configured. Server (4), generated multimodal feature vectors 30 through this, not only the content but also the audio features of calls are included. 6 It is structured to ensure that it is also represented in terms of the server (4), semantic similarities between the same and consecutive calls from the customer measuring the frequency of calls and the temporal continuity between them. by identifying indicators, the usual customer feedback subsequently changes the content. Technically, there are 5 situations where the same issue is raised repeatedly through responses. It is configured to parse. Server (4), LLM-based problem condition Changes in the resolution state throughout the call sequence using tags It is configured to monitor. The server (4) monitors the sequential and repetitive the resolution status in the calls is not progressing, remains unclear, or is inconsistent The repetition of this pattern constitutes a deviation from normal customer behavior and an anomaly. 10 It is structured to ensure that it is evaluated as an indicator. The server (4) displays the multimodal indicators received from the database (3) between calls. time differences, repetition density, level of semantic similarity, resolution status By evaluating the data in terms of continuity and acoustic-emotional form together, many The server (4), 15 is configured to calculate a dimensional anomaly score. By evaluating the customer's past call data, dynamically to define threshold values that can be updated and the anomaly score to these thresholds If the values are exceeded, the relevant call sequence is classified as an anomaly. It is structured accordingly. The server (4) uses these scores based on service type, subject and dynamic and learnable normal behavior created on a time period basis 20 to detect anomalies by comparing them with other models is configured. The server (4) classifies call sequences as anomalies. to verify by comparing data obtained from different communication channels and It is configured to reduce erroneous results. The server (4) calculates Anomaly scores, verification results, and related technical indicators are stored in database 25. (3) storing, using this data, model parameters, threshold updating values and decision-making mechanisms and closed-loop analysis It is configured to continuously improve its performance. The server (4) obtains by enabling the anomaly detection outputs to be transmitted to decision support tools To ensure participation in operational monitoring and evaluation processes 30 It is structured accordingly. 7 Industrial Application of the Invention The system in question (1) is made by calls between the customer and the user. By enabling the multimodal and temporal analysis of interactions, again 5 situations such as missed calls, unresolved problems and operational disruptions This enables automatic detection. Especially in telecommunications. Significant operational capabilities in call centers and industrial areas within their systems. It provides benefits. It eliminates the need for unique and recurring calls associated with the same customer or facility. Instead of being evaluated independently, consecutive calls associated with the same customer 10 sequences (call chains) that are temporal and semantic within specific time windows This allows for analysis on a holistic basis. In this way, independent analysis on the surface is possible. If the visible calls actually point to the same root problem, the system (1) does this early while identifying them at the stage and hidden fault patterns in the call chains the identification of anomalies causing disruptions in the service process 15 It enables the technical differentiation of the search patterns that are formed. Around these basic concepts, the subject of the invention is the “Anomaly Detection System (1)”. It is possible to develop a wide variety of arrangements, and the invention described herein It cannot be limited to examples; it is essentially as stated in the claims. 20
Claims
8 REQUESTS 1. Sequential customer searches based on content, semantic similarity, and temporal orientation. by analyzing them together in terms of continuity, deviations from normal behavior 5. - information on call data from customers, call analysis to view the request and the results of the analyzed call data at least one electronic device that provides a structured interface (2), -contains customer call records, customer data, and temporal data. call sequences, contextual indicator data, anomalies, and anomaly 10 at least one data entry configured to store classification parameters base (3) and -using any remote communication protocol with electronic devices (2) to communicate, to receive call anomaly analysis requests from the user, to collect call and customer information received from call centers and 15 Synchronizing this information over time, in the collected information Pre-processing to eliminate missing or erroneous records and information; subject to preprocessing. Call sequences by grouping the collected data by time and customer. to create, conversation transcripts within the call sequence with LLM Analyzing and categorizing, creating vector representations for each call, 20 Extracting acoustic and emotional characteristics from audio recordings of calls, Multimodal feature vectors by combining features with textual representations. obtaining user intent by classifying the content of incoming calls. Obtaining trust scores, tracking all calls and finding solutions. 25 and calculate the anomaly score, call up the scores and other data obtained. Monitoring and decision support by integrating into central management systems at least one server configured to enable its use in the processes (4) a system characterized by containing (1). 9 2. Smartphone, tablet computer, desktop computer or portable Claim characterized by a computer-like electronic device (2) A system like the one in 1 (1).
3. Connect to the server (4) using any remote communication protocol 5 to establish and facilitate data exchange through this established connection Claim 1 characterized by the configured electronic device (2) or A system like the one in 2 (1).
4. Call logs, user information, and multi-mode 10 entered via the interface. transfer of data components to the server (4) and on these data Anomaly detection generated as a result of LLM-based analyses performed electronic systems configured to display outputs A system (1) as in Claim 3, characterized by device (2).
5. Connecting to the server (4) via a data network such as the Internet. The above is characterized by an electronic device (2) structured to be constructed in accordance with the above. a system like any of the requests (1).
6. Communicating with Server (4) and being managed by Server (4) 20 above characterized by the structured database (3) a system like any of the requests (1).
7. Call records, customer information, time learned via Server (4) stamps, speech transcripts, and audio data, and 25 derived from them semantic outputs, acoustic properties, multimodal representations, inter-call time differences, repetition intensity data, customer emotional data, customer speech rate, energy and intonation information, calls similarity scores between solution state continuity and acoustic-emotional indicators, anomaly scores, caller ID numbers, call sequence 30 relationship information, call sequence indexing data, data synchronization records, customer intent, problem status, main for each call category, subcategory data, parameter configurations, thresholds to enable the storage of values and feedback records above characterized by the structured database (3) a system like any of the requests (1). 5 8. Using any remote communication protocol, electronic device (2) to communicate with and through this communication with electronic devices (2) server (4) configured to exchange data with a 10 as in any of the above characterized claims system (1).
9. Making new data entries into the database (3), database (3) deletion of the registered data in it or within the database (3) Modification of registered data, 15 registered data in database (3) By updating and retrieving data, the database (3) characterized by the server (4) configured to manage a system like any of the above requests (1).
10. 20 for the analysis of call data received via electronic device (2). After the request is received, call records are retrieved from call center infrastructures. Timestamps, customer identification information, automated speech recognition speech transcripts and audio data produced through systems to ensure that it is received and recorded in the database (3) 25 of the above requests characterized by the configured server (4) a system like any other (1).
11. By preprocessing the data recorded in the database (3), common synchronize on the timeline, remove missing or erroneous records and the server (4) configured to make it suitable for analysis 30 11 as in any of the above characterized claims system (1).
12. Grouping pre-processed call data temporally by customer ID. arranging them in order, the time difference between consecutive calls is 5 to calculate and within a predefined sliding time window server configured to include remaining calls in the same call sequence as in any of the above claims characterized by (4). a system (1).
13. Initiating a new call sequence for calls outside the time window. and each call is structured temporally around a specific customer problem. with the server (4) configured to model within the interaction set as in any of the above characterized claims system (1). 15 14. Transcripts of conversations within each call sequence in large languages. By analyzing through model (LLM) based systems (1), the main category, subcategories, customer intent, problem state, and contextual indicators to generate semantic outputs, and from these outputs, to derive semantic 20 for each call to create vector representations and to store these vector representations in the database (3) characterized by the server (4) configured to record a system like any of the above requests (1).
15. Vector representations and calls received from the database (3) 25 to enable semantic comparisons to be carried out from the above requests characterized by the configured server (4) a system like any other (1).
16. In addition to textual analyses, 30 received via electronic device (2) Voice data from calls includes speech rate, voice energy, intonation, and emotion. 12 to extract the acoustic characteristics of the situation and to interpret these characteristics textually combining representations to create multimodal feature vectors for each call. characterized by the server (4) configured to create a system like any of the above requests (1).
17. Through the generated multimodal feature vectors, calls are made only with content. not only is it represented in terms of sound characteristics characterized by the server (4) configured to provide a system like any of the above requests (1).
18. Semantic similarities between the same and consecutive calls from the customer. measuring the frequency of calls and the temporal continuity between them. a regular customer who modifies the content by setting indicators situations where the same problem is repeatedly raised in the feedback loop 15 characterized by the server (4) configured to technically separate a system like any of the above-mentioned requests (1).
19. Throughout the call sequence using LLM-based problem statement labels. It is configured to monitor changes in the solution status. Server (4), the resolution status of the consecutive and repeated calls it followed is 20 failure to progress, remaining uncertain, or repeating inconsistently An indication of deviation and anomaly from normal customer behavior. the server (4) configured to ensure that it is evaluated as such as in any of the above characterized claims system (1). 25 20. Multimodal indicators taken from the database (3), inter-call time differences, repetition density, level of semantic similarity, resolution status by evaluating the data in terms of continuity and acoustic-emotional form together Server 30 configured to calculate a multidimensional anomaly score 13 as in any of the above claims characterized by (4). a system (1).
21. By making assessments based on the customer's past call data, To define dynamically updated threshold values and anomaly 5 If the score exceeds these threshold values, the relevant call sequence will be flagged as an anomaly. characterized by the server (4) configured to be classified as a system like any of the above-mentioned requests (1).
22. These scores are based on a 10-point scale, categorized by service type, topic, and time period. by comparing it with dynamic and learnable normal behavioral models with the server (4) configured to perform anomaly detection as in any of the above characterized claims system (1).
23. Call sequences classified as anomalies from different communication channels verifying by comparing with the obtained data and correcting erroneous conclusions. characterized by the server (4) configured to reduce a system like any of the above requests (1).
24. Calculated anomaly scores, verification results, and related technical data. to store indicators on database (3), using this data model parameters, threshold values, and decision mechanisms to update and continuously improve closed-loop analysis performance The above 25 is characterized by the server (4) configured to do so. a system like any of the requests (1).
25. Communicating the obtained anomaly detection outputs to decision support tools. by contributing to operational monitoring and evaluation processes 30 characterized by the server (4) configured to ensure its presence. a system like any of the above-mentioned requests (1).