Description:FIELD OF THE INVENTION
The present invention relates to customer experience management. More particularly, it pertains to a system and method for aggregating and classifying customer feedback from multiple heterogeneous systems using multi-graph structures and multi-agent artificial intelligence.
BACKGROUND OF THE INVENTION
Several modern organizations collect customer feedback across numerous channels such as email, chat, support tickets, product reviews, surveys, app stores, and social media. Each channel differs structurally and often contains metadata, time-based interactions, sentiment, and product-related context. It becomes a tedious job to manage such heterogeneous feedback data that presents significant challenges, particularly in preserving the contextual relationships and meaning across systems.
In today’s era, conventional systems often normalize feedback into a flat schema, which results in the loss of semantic richness and structural integrity. Other systems rely on keyword-based classification methods that lack scalability, contextual understanding, and adaptability to dynamic inputs. These approaches fail to capture the interrelated meaning and feedback patterns that span across systems, limiting their effectiveness in identifying duplicate issues, classifying feedback accurately, or extracting actionable insights.
Therefore, there exists a need for a system that maintains native structural relationships of each feedback source, extracts relevance across graphs without full normalization, applies AI agents for contextual aggregation and classification, supports a pluggable architecture for new feedback systems and offers explainability and traceability.
Prior Art:
For instance, US10963688B2 discloses systems and methods for classifying customer feedback using sentiment scoring, parts-of-speech filtering, and risk prioritization via term-document matrices. While this disclosure introduces automated feedback analysis, it relies on traditional text-processing techniques and operates on flattened data structures. It does not propose a graph-based representation of feedback, nor does it preserve the native schema or contextual relationships inherent to each system. Moreover, it lacks any mention of a collaborative multi-agent artificial intelligence framework, bridge edge modeling, or explainable cross-system reasoning, all of which are core components of the present invention.
US11200581B2 relates to a multi-client service platform that centralizes contact data across sales, marketing, and service applications using universal contact objects. While it demonstrates multi-source data handling and integration across business functions, it is not designed specifically for the classification or thematic analysis of customer feedback. It does not introduce a semantic or structural model for feedback data, nor does it implement a multi-agent artificial intelligence architecture for deduplication, classification, or feedback motif discovery. The system also lacks mechanisms for establishing cross-platform semantic similarity or providing relevance-ranked insights based on urgency, sentiment, or product features.
DEFINITIONS
The expression “system” used hereinafter in this specification refers to an ecosystem comprising, but is not limited to a system with a user, input and output devices, processing unit, plurality of mobile devices, a mobile device-based application to identify dependencies and relationships between diverse businesses, a visualization platform, and output; and is extended to computing systems like mobile, laptops, computers, PCs, etc.
The expression “input unit” used hereinafter in this specification refers to, but is not limited to, mobile, laptops, computers, PCs, keyboards, mouse, pen drives or drives.
The expression “output unit” used hereinafter in this specification refers to, but is not limited to, an onboard output device, a user interface (UI), a display kit, a local display, a screen, a dashboard, or a visualization platform enabling the user to visualize, observe or analyse any data or scores provided by the system.
The expression “processing unit” refers to, but is not limited to, a processor of at least one computing device that optimizes the system.
The expression “large language model (LLM)” used hereinafter in this specification refers to a type of machine learning model designed for natural language processing tasks such as language generation. LLMs are language models with many parameters, and are trained with self-supervised learning on a vast amount of text.
The expression “Pluggable adapters” used hereinafter in this specification refer to modular components in a software system that can be easily plugged in or swapped to connect with external systems, APIs, databases, services, or tools — without modifying the core logic of the application. They act as interfaces or connectors between your core application and external services. Each adapter knows how to talk to a specific service (e.g., Zendesk, Twitter, a CRM, or a database).
The expression “Transformer LLM (Large Language Model)” used hereinafter in this specification refer to an AI model that uses the transformer architecture to understand, generate, and manipulate human language at scale.
OBJECTS OF THE INVENTION
The primary object of the present invention is to provide a system and method for aggregating, classifying, and analyzing customer feedback from multiple heterogeneous systems using a multi-graph data structure and multi-agent artificial intelligence.
Another object of the invention is to preserve the native structural relationships of each feedback source by generating directed labeled graphs that maintain local schema, entity types, and metadata attributes.
Yet another object of the invention is to enable semantic linkage across feedback systems through bridge edges identified using a similarity function that leverages named entity similarity, embedding scores, and cross-modal metadata.
A further object of the invention is to apply a multi-agent artificial intelligence classification engine that performs local entity classification, cross-graph reasoning, feedback deduplication, and user intent inference using fine-tuned transformer-based agents.
An additional object of the invention is to support a pluggable architecture that allows seamless integration of new feedback systems without requiring a unified schema or global normalization.
A still further object of the invention is to provide a classified, deduplicated, and relevance-ranked output that segments customer feedback by product features, customer segments, urgency, and thematic clusters through APIs or dashboards.
SUMMARY
Before the present invention is described, it is to be understood that the present invention is not limited to specific methodologies and materials described, as these may vary as per the person skilled in the art. It is also to be understood that the terminology used in the description is for the purpose of describing the particular embodiments only and is not intended to limit the scope of the present invention.
The present invention describes a system and method for aggregating and classifying customer feedback from multiple heterogenous systems. The system comprises of feedback ingestion layer module, graph generation module, multi- graph integration module, multi agent classification engine module, and feedback output module. . Each feedback source is parsed into a native graph preserving the data’s local schema, relationships, and attributes . The system uses a multi-graph and multi-agent architecture. Each system's feedback is modeled as a native graph to preserve relationships and structure. Bridge edges are established between graphs using a novel similarity algorithm, and AI agents operate collaboratively to classify and deduplicate feedback across systems. The invention includes arbitration logic, explainable outcomes, and graph motif discovery—offering a scalable and accurate solution for customer feedback intelligence. The system uses novel features of multi-graph fusion, graph relational encoder, multi-agent feedback processing engine, and feedback outcome layer.
According to an aspect of the present invention, each source system generates a native graph. These are not forcefully merged into one global schema. Instead, they are connected via cross-system bridge edges defined by entity similarity (e.g., product, user, sentiment anchor, or time).The proprietary graph encoder computes semantic embeddings of nodes, edge types, and motifs using relation-aware transformers, allowing consistent comparison across graphs. The set of collaborating AI agents is instantiated, each assigned to a graph or subgraph cluster. The classified, deduplicated, and relevance-ranked set of feedback is presented through APIs or dashboards, segmented by product features, customer segments, urgency, and theme.
BRIEF DESCRIPTION OF DRAWINGS
A complete understanding of the present invention may be made by reference to the following detailed description which is to be taken in conjugation with the accompanying drawing. The accompanying drawing, which is incorporated into and constitutes a part of the specification, illustrates one or more embodiments of the present invention and, together with the detailed description, it serves to explain the principles and implementations of the invention.
FIG. 1 illustrates a flowchart of the workflow of the present invention;
FIG. 2: illustrates example of graph nodes and cross-graph bridge edges of the present invention;
FIG.: illustrates multi-agent control flow with feedback loops and arbitration logic of the present invention;
FIG. 4: illustrates feedback classification lifecycle of the present invention.
DETAILED DESCRIPTION OF INVENTION:
Before the present invention is described, it is to be understood that this invention is not limited to methodologies described, as these may vary as per the person skilled in the art. It is also to be understood that the terminology used in the description is for the purpose of describing the particular embodiments only and is not intended to limit the scope of the present invention. Throughout this specification, the word “comprise”, or variations such as “comprises” or “comprising”, will be understood to imply the inclusion of a stated element, integer or step, or group of elements, integers or steps, but not the exclusion of any other element, integer or step, or group of elements, integers or steps. The use of the expression “at least” or “at least one” suggests the use of one or more elements or ingredients or quantities, as the use may be in the embodiment of the invention to achieve one or more of the desired objects or results. Various embodiments of the present invention are described below. It is, however, noted that the present invention is not limited to these embodiments, but rather the intention is that modifications that are apparent are also included.
The present invention describes a system and method for aggregating and classifying customer feedback as individual knowledge graphs from multiple source systems. Each feedback source is parsed into a native graph preserving the data’s local schema, relationships, and attributes. Some examples of customer feedback sources are customer service and support platforms, social media platforms, app marketplaces, or business software platforms. Some examples are: Zendesk, Twitter, Play Store, CRM, etc.
According to the embodiment of the present invention, the system comprises of an input unit , a processing unit and output unit , wherein the processing unit further comprises of feedback ingestion layer module, graph generation module, multi- graph integration module, multi agent classification engine module, and feedback output module. In the feedback ingestion layer module, feedback from multiple systems is ingested through pluggable adapters. Each feedback is timestamped, tagged with system-specific metadata, and optionally linked to customer profile data. In the graph generation module, each customer feedback source system (S) is converted into a directed labeled graph (G). Each graph consists of nodes (V) that represents the entities or objects in the system , such as customer, ticket, app version; Edges ( E) that represents connections or relationships between the nodes, such as reported by (ticket- customer) , affects (ticket- app version), mentions (ticket- product); and attributes that represent properties or metadata associated with either nodes or edges for example sentiment, timestamp, severity, status, location, etc.
According to the embodiment of the present invention, as described in FIG. 3, in multi- graph integration module, the system preserves separate graphs and connects them through bridge edges. A bridge edge is a semantic link between two nodes from different graphs (v_a and v_b) that are likely to refer to the same or related entities. The system uses a similarity function- sim(v_a, v_b). This function measures how similar two nodes (from different graphs) are, using the following:
• Named Entity Similarity: It compares names of entities (e.g., person names, product names, cities). Example: “New York” in one graph vs. "NYC" in another → high similarity. It uses NLP techniques like entity resolution and normalizations.
• Embedding Cosine Scores: Each node is represented as a vector embedding. Cosine similarity is then used to measure closeness between these vectors. It captures semantic similarity, even if the exact words don’t match. Example: "angry customer" and "user frustration" might be close in embedding space.
• Cross-Modal Metadata Matching: It looks at non-textual, structured metadata that spans different modalities (e.g., time, location, product category). Example: A complaint from a certain geo-location that matches with a device failure log from the same region and time period, even though they come from entirely different data sources.
Using this module avoids schema flattening as it doesn’t try to force-fit all data into a common structure. Each graph retains its own schema and structure, while enabling cross-system reasoning through a multi-graph approach. It also maintains the original meaning, granularity, and domain-specific nuance of each graph, thus preserving high fidelity. It also includes a motif mining mechanism that detects root causes and complaint types via recurring graph patterns across feedback systems.
According to the embodiment of the present invention, as described in FIG. 2 and FIG. 4 in the multi agent classification engine module, different agents operate in parallel. The local agents tag nodes within each directed labeled graph (G). The bridge agents operate on bridge edges to resolve cross-graph equivalence or causal relationships. Consensus Agents aggregate multiple viewpoints (votes, confidence) using arbitration logic. Each agent uses a fine-tuned transformer LLM and memory is maintained using vector stores linked to graphs. Finally, the output module generates a unified feedback graph with semantic clusters, classification tags (bug, request, praise, urgency), topic modelling via motif discovery, anomaly detection for surges. The unified feedback graph is a classified, deduplicated, and relevance-ranked set of feedback that is presented through APIs or dashboards, segmented by product features, customer segments, urgency, and theme.
According to the embodiment of the present invention, the present system and method for aggregating, classifying, and analyzing customer feedback uses a multi-graph and multi-agent architecture. Each system's feedback is modeled as a native graph to preserve relationships and structure. Bridge edges are established between graphs using a novel similarity algorithm, and AI agents operate collaboratively to classify and deduplicate feedback across systems. The invention includes arbitration logic, explainable outcomes, and graph motif discovery—offering a scalable and accurate solution for customer feedback intelligence.
According to the embodiment of the present invention, the system uses novel features of multi-graph fusion, graph relational encoder, multi-agent feedback processing engine, and feedback outcome layer.
• Multi-Graph Fusion: Each source system generates a native graph (G1, G2, G3...Gn). These are not forcefully merged into one global schema. Instead, they are connected via cross-system bridge edges defined by entity similarity (e.g., product, user, sentiment anchor, or time).
• Graph Relational Encoder: A proprietary graph encoder computes semantic embeddings of nodes, edge types, and motifs using relation-aware transformers, allowing consistent comparison across graphs.
• Multi-Agent Feedback Processing Engine: A set of collaborating AI agents is instantiated, each assigned to a graph or subgraph cluster. These agents perform:
1) Local entity classification (bug, request, sentiment),
2) Cross-graph reasoning (is this Play Store bug same as the Zendesk ticket?),
3) Feedback de-duplication, and
4) User intent inference.
• Feedback Outcome Layer: A classified, deduplicated, and relevance-ranked set of feedback is presented through APIs or dashboards, segmented by product features, customer segments, urgency, and theme.

According to the embodiment of the present invention, the method for aggregating and classifying customer feedback from multiple heterogenous systems as described in Fig. 1 includes the steps of:
• ingesting feedback from multiple heterogenous systems by feedback ingestion layer module;
• constructing a native labeled graph for each system by the graph generation system module where each graph has nodes, edges and attributes;
• identifying semantic bridge edges across graphs using multi-graph integration module that uses a relation-aware transformer similarity function applied on named entities and metadata;
• applying multi-agent classification engine module to classify, deduplicate, and cluster feedback and achieving agent consensus using arbitration logic based on confidence scores, historical accuracy, and voting;
• generating output of classified, deduplicated, and relevance-ranked set of feedback by the output module.
According to the embodiment of the present invention, the advantages of the present system and method for aggregating, classifying, and analyzing customer feedback from multiple heterogeneous systems include:
1.Bridge Edge Graph Retention Model (BEGRM): Instead of a merged global graph, native graphs are retained and only key entity alignments are bridged. This approach preserves structure and improves explainability—unlike vector-only similarity models.
2. Multi-Agent Arbitration Engine (MAAE): Distinct LLM-based agents classify feedback with delegated roles. A consensus arbiter agent selects the most confident result or flags ambiguity for human review—offering a scalable yet explainable pipeline.
3. Cross-Graph Motif Mining (CGMM): Feedback motifs like “crash after update” or “login timeout after 3 days” are identified not from text alone but from recurrent graph patterns across systems.
, Claims:We claim,
1. A system and method for aggregating and classifying customer feedback from multiple heterogenous systems
characterized in that
the system comprises of an input unit , a processing unit and output unit , wherein the processing unit further comprises of feedback ingestion layer module, graph generation module, multi- graph integration module, multi agent classification engine module, and feedback output module; and
the method for aggregating and classifying customer feedback includes the steps of:
• ingesting feedback from multiple heterogenous systems by feedback ingestion layer module;
• constructing a native labelled graph for each system by the graph generation system module where each graph has nodes, edges and attributes;
• identifying semantic bridge edges across graphs using multi-graph integration module that uses a relation-aware transformer similarity function applied on named entities and metadata;
• applying multi-agent classification engine module to classify, deduplicate, and cluster feedback and achieving agent consensus using arbitration logic based on confidence scores, historical accuracy, and voting;
• generating output of classified, deduplicated, and relevance-ranked set of feedback by the output module.

2. The system and method as claimed in claim 1, wherein the systems for customer feedback sources are customer service and support platforms, social media platforms, app marketplaces, or business software platforms.

3. The system and method as claimed in claim 1, wherein in the feedback ingestion layer module, feedback from multiple systems is ingested through pluggable adapters and each feedback is timestamped, tagged with system-specific metadata, and optionally linked to customer profile data.

4. The system and method as claimed in claim 1, wherein in the graph generation module, each customer feedback source system is converted into a directed labeled graph such that each graph consists of nodes that represents the entities or objects in the system ; edges that represents connections or relationships between the nodes; and attributes that represent properties or metadata associated with either nodes or edges.

5. The system and method as claimed in claim 1, wherein in multi- graph integration module, the system preserves separate graphs and connects them through bridge edges which is a semantic link between two nodes from different graphs that are likely to refer to the same or related entities.

6. The system and method as claimed in claim 1, wherein the system uses a similarity function that measures how similar two nodes from different graphs are, using Named Entity Similarity, Embedding Cosine Scores and Cross-Modal Metadata Matching.

7. The system and method as claimed in claim 1, wherein each graph retains its own schema and structure; while enabling cross-system reasoning through a multi-graph approach and it also maintains the original meaning, granularity, and domain-specific nuance of each graph, thus preserving high fidelity.

8. The system and method as claimed in claim 1, wherein in the multi agent classification engine module, different agents operate in parallel that include local agents that tag nodes within each directed labeled graph; The bridge agents that operate on bridge edges to resolve cross-graph equivalence or causal relationships and consensus agents that aggregate multiple viewpoints (votes, confidence) using arbitration logic such that each agent uses a fine-tuned transformer LLM and memory is maintained using vector stores linked to graphs.

9. The system and method as claimed in claim 1, wherein the output module generates a unified feedback graph with semantic clusters, classification tags, topic modelling via motif discovery, anomaly detection for surges and the unified feedback graph is a classified, deduplicated, and relevance-ranked set of feedback that is presented through APIs or dashboards, segmented by product features, customer segments, urgency, and theme.