A method, device and storage medium for constructing a consumer knowledge graph

Abstract

The application discloses a kind of construction methods, device and equipment of consumer knowledge graph and storage medium. Including: obtaining the consumer insight data of target data source, target data source includes at least one;Based on consumer insight data, obtain consumer to-do task;Based on consumer to-do task, generate genesis consumer population node corresponding to consumer to-do task;Based on genesis consumer population node, construct consumer knowledge graph.Through obtaining multi-source consumer insight data, market consumption information can be comprehensively captured, and rich data basis is provided for subsequent analysis.Using the above technical solution, the deep needs of consumers can be accurately mined, breaking through the limitations of traditional focus on behavior data, emerging consumer groups can be automatically discovered, the problem that manual definition is difficult to mine new fields can be solved, a structured association network with consumer demand as the core can be built, accurate population knowledge support is provided for enterprise decision-making, and the graph landing from demand to population is realized.

Description

Technical Field

[0001] This invention relates to the field of knowledge graphs, and more particularly to a method, apparatus, device, and storage medium for constructing a consumer knowledge graph. Background Technology

[0002] Against the backdrop of the rapid development of the digital consumer market, consumer demand is becoming increasingly dynamic and diversified. Enterprises urgently need knowledge support to accurately grasp the evolutionary patterns and deeper needs of different consumer groups. Knowledge graphs, as a core tool for structured storage of entities and relationships, are widely used in user modeling. Based on this, consumer knowledge graphs have emerged. Consumer knowledge graphs process consumer-related data in a structured, relational, and intelligent manner, integrating multi-source data to construct a network of relationships related to consumer characteristics. This helps enterprises gain a deeper understanding of consumers and provides a basis for market decisions and product innovation.

[0003] Current related technologies mainly fall into four categories: Static user profiling systems build tag systems based on demographic features and rely on regular manual updates. Traditional knowledge graph construction systems require manual definition of the ontology structure, extract entity relationships through rules or supervised learning, and update them in batches. Single-agent large language model applications rely on a single large language model to complete question-answering or generation tasks and lack self-verification mechanisms. Traditional recommendation systems focus on behavioral data for user modeling, building interest models through algorithms such as collaborative filtering.

[0004] In existing technologies, static profiles cannot track the evolution of the user lifecycle and are difficult to reflect real-time market changes. Traditional knowledge graphs rely on manual definition, have insufficient ability to discover new domains, lag in knowledge updates, are prone to illusions due to single agents, lack credibility and evidence support for outputs, have poor interpretability, and have rigid user segmentation standards, making it impossible to adaptively segment emerging groups. At the same time, existing technologies generally lack modeling of consumers' emotional and social motivations, and have high human involvement and maintenance costs, making it difficult to form continuously accumulated enterprise-level consumer knowledge assets and failing to meet the needs of enterprises for precise and long-term decision-making. Summary of the Invention

[0005] This invention provides a method, apparatus, device, and storage medium for constructing a consumer knowledge graph, which solves the technical problems of traditional user profiling and knowledge graph systems being static and fixed, reliant on manual intervention, unreliable single-agent output, unable to adaptively discover new domains, and lagging knowledge updates.

[0006] This invention provides a method for constructing a consumer knowledge graph, the method comprising: Obtain consumer insight data from target data sources, including at least one such data source; Based on consumer insight data, obtain consumers' to-do tasks; Based on the consumer's to-do tasks, generate the corresponding Genesis Consumer Group node for the consumer's to-do tasks. A consumer knowledge graph is constructed based on the genesis consumer group nodes.

[0007] Optionally, consumer insight data includes at least one of macro trend data, search trend data, and user-generated content (UGC) data; obtaining consumer to-do tasks based on consumer insight data includes: identifying target trend signals in consumer insight data that meet predetermined change trends; identifying anomalous phenomena that differ from historical phenomena based on target trend signals; and performing cluster analysis on consumer insight data related to anomalous phenomena to identify emerging consumer to-do tasks.

[0008] Optionally, the predetermined trend of change includes at least one of the following: the number of mentions in the predetermined time window increases by more than a first specified threshold compared to the number of mentions in the previous time window; it shows a monotonically increasing trend over a first specified number of predetermined time periods; it spreads from mentions in a single data source to mentions in at least a second specified number of data sources; it appears for the first time or changes from mentions in a predetermined low-frequency range to mentions in a predetermined high-frequency range.

[0009] Optionally, generating the Genesis Consumer Group node corresponding to the consumer to-do task includes: defining the name, definition, coverage, and characteristics of the emerging consumer group corresponding to the emerging consumer to-do task; and generating a structured Genesis Consumer Group node based on the name, definition, coverage, and characteristics of the emerging consumer group.

[0010] Optionally, after constructing the consumer knowledge graph, the method further includes: conducting vertical evolution prediction and / or horizontal segmentation hypotheses on the generative consumer group nodes or standard consumer group nodes already in the database to form new node hypotheses; supplementing the task attribute hypotheses for the new node hypotheses; iteratively verifying the new node hypotheses that supplement the task attribute hypotheses through multi-agent adversarial debate, and transforming the verified new node hypotheses into standard consumer group nodes and writing them into the knowledge graph.

[0011] Optionally, longitudinal evolution predictions are performed on the genealogy consumer group node or the standard consumer group node already in the database to form new node hypotheses, including: using the genealogy consumer group node or the standard consumer group node already in the database as the original node, and performing at least one of the following predictions on the original node: lifecycle evolution prediction, event-triggered evolution prediction, and trend-driven evolution prediction, to form longitudinal evolution node hypotheses; and / or, horizontal segmentation hypotheses are performed on the genealogy consumer group node or the standard consumer group node already in the database to form new node hypotheses, including: Using the original consumer group node or the standard consumer group node already in the database as the original node, conduct at least one of the following analyses on the original node: behavioral difference analysis, tag combination analysis, and demand analysis, in order to form the hypothesis of horizontal segmentation node.

[0012] Optionally, after forming a new node hypothesis and before supplementing the task attribute hypothesis for the new node hypothesis, the method further includes: validating the new node hypothesis, wherein the validity verification includes at least one of data stock verification, relation legality verification and graph consistency verification; supplementing the task attribute hypothesis for the new node hypothesis includes: supplementing the task attribute hypothesis for the new node hypothesis that has passed the validity verification.

[0013] Optionally, the new node hypothesis is validated using existing data, including: counting the number of discussions about the new node hypothesis; if the number of discussions exceeds a preset number, the data validation is deemed successful; and / or, the new node hypothesis is validated for relational legitimacy, including: determining the relation type of the new node hypothesis; if the relation type is unique and meets the preset validation criteria corresponding to the relation type, the relational legitimacy validation is deemed successful, wherein the relation type includes evolutionary relations and subdivision relations; and the new node hypothesis is validated for graph consistency, including: determining whether there is a logical contradiction between the new node hypothesis and existing nodes in the knowledge graph; if there is no logical contradiction, the graph consistency validation is deemed successful.

[0014] Optionally, the task attribute hypothesis is supplemented for the new node hypothesis, including: parsing the new node hypothesis to obtain the node name, definition and known features of the new node hypothesis, and integrating them into the target population features; retrieving similar nodes in the consumer knowledge graph based on the target population features, and extracting the task attributes of the similar nodes as reference task attributes; and generating task attribute hypotheses through the hypothesis generation agent based on the reference attributes, wherein the task attribute hypotheses include at least one of functional task attribute hypotheses, emotional task attribute hypotheses and social task attribute hypotheses.

[0015] Optionally, the new node hypothesis that completes the task attribute hypothesis is iteratively verified through multi-agent adversarial debate, including: transforming the task attribute hypothesis into search keywords through an evidence retrieval agent, retrieving consumer insight data based on the search keywords, filtering relevant evidence, and generating an evidence report; sending the evidence report to supporter agents and critic agents respectively for debate, wherein the supporter agents are used to form supporting arguments, and the critic agents are used to form challenging arguments; evaluating the supporting arguments and challenging arguments through a reviewer agent, generating a quality score, and iteratively verifying the hypothesis based on the quality score.

[0016] Optionally, iterative verification is performed based on the quality score, including: obtaining a preset score threshold and a preset debate round, and determining the current debate round; when the quality score reaches the preset score threshold for two consecutive rounds, the new node hypothesis is confirmed to have passed verification; when the quality score does not reach the preset score threshold for two consecutive rounds and the current debate round does not exceed the preset debate round, the debate focus is determined by the evidence retrieval agent, relevant evidence is supplemented based on the debate focus, and the evidence report is updated. The updated evidence report is sent to the supporter agent and the critic agent for a re-debate; when the quality score does not reach the preset score threshold for two consecutive rounds and the current debate round reaches the preset debate round, the cumulative score is calculated, and a final decision is made based on the cumulative score.

[0017] According to another aspect of the present invention, an apparatus for constructing a consumer knowledge graph is provided, comprising: The consumer insight data acquisition module is used to acquire consumer insight data from target data sources, including at least one target data source. The task acquisition module is used to acquire consumers' tasks based on consumer insight data. The Genesis Node Generation Module is used to generate Genesis Consumer Group Nodes corresponding to Consumer To-Do Tasks based on Consumer To-Do Tasks. The Consumer Knowledge Graph Construction Module is used to build a consumer knowledge graph based on the genesis consumer group nodes.

[0018] According to another aspect of the present invention, an electronic device is provided, the electronic device comprising: At least one processor; and A memory communicatively connected to the at least one processor; wherein, The memory stores a computer program that can be executed by the at least one processor, the computer program being executed by the at least one processor to enable the at least one processor to perform the consumer knowledge graph construction method according to any embodiment of the present invention.

[0019] According to another aspect of the present invention, a computer-readable storage medium is provided, the computer-readable storage medium storing computer instructions for causing a processor to execute and implement the consumer knowledge graph construction method according to any embodiment of the present invention.

[0020] The technical solution of this invention, by acquiring multi-source consumer insight data, can comprehensively capture market consumption information, providing a rich data foundation for subsequent analysis. By acquiring consumers' to-do tasks, it can accurately uncover consumers' deep needs, breaking through the limitations of traditional methods that only focus on behavioral data. It can automatically discover emerging consumer groups, solving the problem that manual definition is insufficient for uncovering new areas. By constructing a consumer knowledge graph, a structured network of connections centered on consumer needs can be built, providing precise demographic knowledge support for enterprise decision-making and realizing the graph-based implementation from needs to demographics.

[0021] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of the present invention, nor is it intended to limit the scope of the invention. Other features of the invention will become readily apparent from the following description. Attached Figure Description

[0022] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0023] Figure 1 This is a flowchart of a method for constructing a consumer knowledge graph according to Embodiment 1 of the present invention; Figure 2 This is a flowchart of another method for constructing a consumer knowledge graph according to Embodiment 2 of the present invention; Figure 3 This is a flowchart of another method for constructing a consumer knowledge graph according to Embodiment 2 of the present invention; Figure 4 This is a schematic diagram of a consumer knowledge graph construction device according to Embodiment 3 of the present invention; Figure 5 This is a schematic diagram of the structure of an electronic device that implements a consumer knowledge graph construction method according to an embodiment of the present invention. Detailed Implementation

[0024] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0025] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0026] Example 1 Figure 1 This is a flowchart illustrating a method for constructing a consumer knowledge graph according to Embodiment 1 of the present invention. This embodiment is applicable to consumer group identification scenarios. The method can be executed by a consumer knowledge graph construction device, which can be implemented in hardware and / or software and can be configured in a computer controller. Figure 1 As shown, the method includes: S110. Obtain consumer insight data from a target data source, which may include at least one such data source.

[0027] Among them, the target data source is the specific source carrier and content collection of consumer insight data, which is the original data source from which consumer insight data is extracted.

[0028] Optionally, consumer insight data may include at least one of macro trend data, search trend data, and user-generated content (UGC) data.

[0029] Macroeconomic trend data, reflecting the overall development trend of the industry, overall market changes, and policy guidance, is a comprehensive data source and a fundamental data type for uncovering macroeconomic patterns in consumption. It includes industry reports, market research and analysis materials, policy documents, and other related content, capturing signals of major changes in the consumer market at a holistic level. Search trend data refers to data on the search popularity, frequency, and trends of different keywords on various search engine platforms, directly reflecting changes in consumers' current focus and demand preferences. User-generated content (UGC) refers to various unstructured data content created, shared, and published by consumers themselves. This includes consumer discussions on social media, user posts in emerging topic communities, product reviews on various platforms, and text signals related to user behavior.

[0030] S120. Based on consumer insight data, obtain consumers' to-do tasks.

[0031] Among them, consumer to-do tasks refer to the specific tasks that consumers want to complete in order to achieve a certain goal in a specific scenario. They are a concrete expression of consumers' core needs and are divided into three categories: functional, emotional, and social.

[0032] Optionally, obtaining consumer to-do tasks based on consumer insight data includes: identifying target trend signals in the consumer insight data that meet predetermined change trends; identifying anomalous phenomena that differ from historical phenomena based on the target trend signals; and performing cluster analysis on the consumer insight data related to the anomalous phenomena to identify emerging consumer to-do tasks.

[0033] Among them, target trend signals refer to consumer market trend information identified from consumer insight data that meets pre-set quantitative and qualitative change standards. Anomalies refer to new consumer phenomena that differ significantly from historical patterns and performance in the consumer market, as derived from the analysis of target trend signals. Emerging consumer tasks refer to previously unexplored or entirely new consumer tasks discovered through cluster analysis of consumer insight data related to anomalies.

[0034] First, the system sets clear predetermined change standards from both quantitative and qualitative dimensions. It conducts a comprehensive analysis of macro trend data, search trend data, and user-generated content (UGC) data. Quantitatively, it detects whether the word frequency and mentions of target keywords or entities have reached the threshold for a month-on-month surge, such as a month-on-month increase of more than 200% in the current time window compared to the previous window. It also verifies whether the data meets the momentum sustainability requirement, i.e., showing a significant monotonically increasing trend through the Mann-Kendall trend test. It also checks the diffusion index of the signal to confirm whether the data source types mentioning the signal have reached a specified number across platforms. Qualitatively, it determines whether the relevant words are zero samples appearing for the first time or have changed from low frequency to high frequency, and whether they belong to the general stop word list. Only information that meets both quantitative and qualitative predetermined standards will be identified as the target trend signal.

[0035] Secondly, the system compares and analyzes the extracted target trend signals with historical data patterns and performance in the consumer market. It uses three criteria to detect fundamental differences: First, the semantic drift criterion calculates the cosine distance between the word vectors of the target entity in the current and historical time periods; a result greater than 0 indicates a fundamental change in meaning or application scenario. Second, the sentiment polarity reversal criterion checks whether the sentiment score of the target entity crosses the 0 axis within a short period with a change exceeding 0.5, indicating a significant shift in sentiment. Third, the co-occurrence network structure mutation criterion checks whether the weights of old strong connections decrease by more than 50%, while the weights of newly established connections exceed a threshold. Meeting any of these criteria confirms that the target trend signal corresponds to an anomaly significantly different from historical phenomena.

[0036] Finally, the system preprocesses all consumer insight data related to anomalies. Using pre-trained large models such as Bidirectional Encoder Representations from Transformers (BERT) and Robustly Optimized BERT Pretraining Approach (RoBERTa), unstructured data such as user comments, behavior logs, and text signals are transformed into high-dimensional dense vectors. Then, the high-dimensional vectors are reduced in dimensionality using Uniform Manifold Approximation and Projection (UMAP) and t-Distributed Stochastic Neighbor Embedding (t-SNE) algorithms, preserving the local topological structure of the data while removing sparse noise. Finally, a Hierarchical Density-Based Spatial Clustering of Applications algorithm is employed. The density-based clustering algorithm with Noise (HDBSCAN) clusters the dimensionality-reduced data. This algorithm does not require pre-specifying the number of clusters and can automatically identify and remove noise points, grouping data with similar features into the same cluster. This helps to uncover emerging consumer groups with blurred boundaries. Then, features are extracted from each cluster. The core feature labels within the cluster are obtained through class-based term frequency-inverse document frequency (c-TF-IDF) variant analysis. Combined with a large model, generative summarization is performed to clarify the values, behavioral characteristics, and core needs of this emerging group. Finally, the core needs are concretized into three categories: functional, emotional, and social, which are the to-do tasks of emerging consumers.

[0037] Optionally, the predetermined trend of change includes at least one of the following: the number of mentions in the predetermined time window increases by more than a first specified threshold compared to the number of mentions in the previous time window; it shows a monotonically increasing trend over a first specified number of predetermined time periods; it spreads from mentions in a single data source to mentions in at least a second specified number of data sources; it appears for the first time or changes from mentions in a predetermined low-frequency range to mentions in a predetermined high-frequency range.

[0038] Among them, the predetermined trend is the basis for identifying target trend signals from consumer insight data. It sets up four judgment criteria around the data's popularity, change pattern, diffusion range and occurrence characteristics. Each criterion has clear quantitative or qualitative rules.

[0039] It should be noted that various standards can be used individually or in combination as the basis for judgment, and meeting at least one of them is sufficient to determine that the data conforms to the predetermined trend of change.

[0040] Specifically, the first type of predetermined trend is defined as a growth in the number of mentions within a predetermined time window exceeding a first specified threshold compared to the number of mentions in the previous time window. This standard is determined from the perspective of short-term data growth. It involves statistically analyzing the frequency or number of mentions of target keywords, entities, or consumption-related signals within a set time window and comparing it with the corresponding data in the previous time window of the same dimension. When the calculated growth rate exceeds the first specified threshold, it meets the criteria for this trend. The predetermined time window can be configured to different durations such as 24 hours, 7 days, or a quarter, depending on the actual analysis needs. The first specified threshold can be configured to a specific percentage such as 200%, thereby capturing the short-term surge characteristics of the data and determining whether there is a sudden trend of consumer attention. The second type of predetermined trend is characterized by a monotonically increasing trend over a first specified number of predetermined time periods. This standard is judged from the perspective of the sustainability of data growth. The core of this standard is to use the Mann-Kendall trend test to statistically verify the target data. When the test result meets the significance requirement of P<0.05, it can be determined that the data shows a significant monotonically increasing trend over a first specified number of predetermined time periods. Here, the first specified number is the number of pre-set time periods, and the predetermined time periods are fixed time units. This standard can effectively distinguish between one-off, pulse-like data fluctuations and true trends with sustained growth momentum, avoiding misjudging occasional traffic fluctuations as valuable consumer trend signals. The third type of predetermined trend is when a signal spreads from a single data source to at least a second specified number of data sources. This standard is judged from the perspective of the scope of data dissemination. It is determined by statistically analyzing the number of data source types that mention the target consumer signal. When the signal's dissemination channels gradually spread from the initial single data source to at least a second specified number of different types of data sources, it meets this trend. These data sources include different types such as industry reports, social media platforms, search engines, and emerging topic communities. The second specified number can be configured to be three or more. This standard can determine the breadth and influence of the consumer signal's dissemination. Only signals that achieve cross-platform and cross-type data source dissemination represent a certain degree of market penetration and research value.The fourth category of predetermined trends refers to the first appearance of a term or a shift from a predetermined low-frequency range to a predetermined high-frequency range. This standard combines qualitative and quantitative judgment based on the appearance and frequency change characteristics of the data, and is divided into two scenarios. The first scenario is that the target consumption-related terms or signals appear for the first time and have no relevant records in past consumer insight data, which is a zero-sample signal. The second scenario is that the target signal was originally mentioned in a predetermined low-frequency range, and its frequency or mention volume has been at a low level for a long time. Later, it changes to being mentioned in a predetermined high-frequency range, with a significant increase in frequency. At the same time, this standard also includes a constraint condition: the signal terms that are judged to conform to the trend cannot appear in the general stop word list, so as to avoid including general terms without actual consumer significance in the trend signal range. This standard can effectively identify newly emerging consumer concepts in the market, or consumer trends that have shifted from niche attention to mainstream attention.

[0041] S130. Based on the consumer's to-do tasks, generate the corresponding Genesis Consumer Group node.

[0042] Among them, the Genesis Consumer Group Node is the initial seed node for building the consumer knowledge graph. Based on consumer insight data and the definition of consumer to-do tasks, it has clear group characteristics and boundaries.

[0043] Optionally, generating the Genesis Consumer Group node corresponding to the consumer to-do task includes: defining the name, definition, coverage, and characteristics of the emerging consumer group corresponding to the emerging consumer to-do task; and generating a structured Genesis Consumer Group node based on the name, definition, coverage, and characteristics of the emerging consumer group.

[0044] Specifically, based on emerging consumer tasks, the system clarifies the name, definition, coverage, and characteristics of the emerging consumer group corresponding to those tasks. This involves using the attributes and related data of the emerging consumer tasks, combined with domain boundary definition techniques, to comprehensively extract features and define the scope of the emerging consumer group. The system integrates all clustering analysis results, consumer insight data, and feature tags related to the emerging consumer tasks. Using a c-TF-IDF variant analysis method, it treats the cluster corresponding to the emerging consumer group as a whole, calculating the core words that are high-frequency within the cluster and low-frequency in other clusters, thus determining the core feature tags of the group. Then, the core feature tags, core cluster samples, and the specific content of the emerging consumer tasks are input into a large language model. Through generative summarization, combined with manually configurable prompts, the model generates candidate names for the emerging consumer group based on the correlation between needs and features. Simultaneously, it clarifies the core definition of the group, requiring a precise match to the needs and characteristics reflected in the emerging consumer tasks, clearly describing the group's identity, consumption demands, and other core attributes. When defining the scope, a centroid and radius calculation method is used. First, the geometric center of all data points within the corresponding cluster of the group is calculated. Then, the distance distribution from all points within the cluster to the geometric center is calculated. The distance of the 90th percentile is taken as the effective boundary radius of the group. This excludes loose data cases at the edge, clarifies the core coverage of the group, and defines which consumer groups belong to the group and which need to be excluded. At the same time, the scope boundary is further refined in combination with the applicable scenarios of emerging consumer tasks to ensure that the scope definition matches the task requirements. When extracting group features, in addition to the core feature labels extracted by the algorithm, the three types of attributes of emerging consumer tasks can be used as a supplement. The actual consumption behavior characteristics of the group are extracted from functional tasks, the psychological and emotional characteristics of the group are extracted from emotional tasks, and the social identity characteristics of the group are extracted from social tasks. At the same time, the concrete characteristics of the group such as skin type, age, and usage scenarios in the consumer insight data are integrated to form a complete feature system of this emerging consumer group.

[0045] Furthermore, the system will structure the previously defined unstructured group information according to a preset standard data format, forming a genealogy consumer group node that can be directly incorporated into the consumer task knowledge graph. First, the system will assign a unique node identifier to this genealogy consumer group node. Then, the system will fill in the previously identified emerging consumer group's standard name, precise definition, clear coverage description, and complete characteristic system according to preset field requirements. Simultaneously, key metadata information will be added to the structured node, including the discovery source of the genealogy node, clarifying whether it was mined from macro trend data, search trend data, user-generated content (UGC) data, or the result of multi-source data fusion; the node's creation time will also be recorded to form complete node traceability information, facilitating subsequent graph updates and verification. Finally, the system will integrate all the above structured information according to the preset general data format to form the Genesis Consumer Group Node, which includes core fields such as node identifier, node name, definition, coverage, initial characteristics, discovery source, and creation time. As a seed node for a new field, this node has a standardized structure and complete information, and can directly provide input for the subsequent node expansion and reasoning module, supporting the subsequent expansion and evolution of the consumer knowledge graph.

[0046] S140. Construct a consumer knowledge graph based on the genesis consumer group nodes.

[0047] Among them, a knowledge graph is a knowledge base that stores entities and their relationships in a graph structure, while a consumer knowledge graph is a dynamic knowledge base specifically for consumer tasks that uses a graph structure as its core storage format.

[0048] Specifically, the system will input the generated structured genealogy consumer group nodes into the consumer knowledge graph's group node library according to the preset standard data format. This node contains information such as unique identifier, name, definition, coverage, initial characteristics, discovery source, and creation time. It is the starting point for all subsequent expansion work of the graph. At the same time, it reserves an interface for attribute completion and relationship connection in the graph for this node, preparing it for establishing associations with other nodes and supplementing task attributes in the future.

[0049] The technical solution of this invention, by acquiring multi-source consumer insight data, can comprehensively capture market consumption information, providing a rich data foundation for subsequent analysis. By acquiring consumers' to-do tasks, it can accurately uncover consumers' deep needs, breaking through the limitations of traditional methods that only focus on behavioral data. It can automatically discover emerging consumer groups, solving the problem that manual definition is insufficient for uncovering new areas. By constructing a consumer knowledge graph, a structured network of connections centered on consumer needs can be built, providing precise demographic knowledge support for enterprise decision-making and realizing the graph-based implementation from needs to demographics.

[0050] Example 2 Figure 2 This is a flowchart of a method for constructing a consumer knowledge graph according to Embodiment 2 of the present invention. This embodiment adds a dynamic evolution process of the consumer knowledge graph based on Embodiment 1. Figure 2 As shown, the method includes: S210. Conduct longitudinal evolution prediction and / or horizontal segmentation hypotheses for the generative consumer group node or the standard consumer group node already included in the database to form new node hypotheses.

[0051] Vertical evolution prediction refers to the process of predicting the direction of state changes of nodes over time, due to external events, or market trends, from the dimensions of life cycle, event triggers, and trend-driven factors. This process generates new vertical nodes with temporal or developmental stage connections, and is a method for expanding the user base nodes in the time dimension. It can capture the natural development of consumer groups and identity changes caused by external factors. Horizontal segmentation hypothesis refers to the process of mining subgroups with different characteristics within nodes from the dimensions of behavioral differences, tag combinations, and user-generated content themes. This process generates new horizontal nodes belonging to subsets of the original nodes, discovering subgroups with unique attributes within the original user base and achieving refined expansion of user nodes in terms of dimensions. New node hypothesis refers to the preliminary consumer group node prototypes generated by conducting vertical evolution prediction and / or horizontal segmentation hypothesis on the original consumer group nodes or standard consumer group nodes already in the database. These prototypes, which have not yet undergone structural verification and attribute completion, include the node's initial name, definition, and characteristic assumptions.

[0052] Optionally, vertical evolution prediction can be performed on the genealogy consumer group node or the standard consumer group node already included in the database to form new node hypotheses. This includes: taking the genealogy consumer group node or the standard consumer group node already included in the database as the original node, and performing at least one of the following predictions on the original node: life cycle evolution prediction, event-triggered evolution prediction, and trend-driven evolution prediction, to form vertical evolution node hypotheses; and / or, horizontal segmentation hypotheses can be performed on the genealogy consumer group node or the standard consumer group node already included in the database to form new node hypotheses. This includes: taking the genealogy consumer group node or the standard consumer group node already included in the database as the original node, and performing at least one of the following analyses on the original node: behavioral difference analysis, tag combination analysis, and demand analysis, to form horizontal segmentation node hypotheses.

[0053] It should be noted that the longitudinal evolution prediction focuses on the state changes of population nodes in the time and development dimensions, while the horizontal segmentation hypothesis focuses on the refined group mining within population nodes. The two expansion methods can be carried out separately or simultaneously.

[0054] Specifically, the system uses either the founding consumer group node or the standard consumer group node already in the database as the origin node. Through at least one of the following prediction methods—lifecycle evolution prediction, event-triggered evolution prediction, and trend-driven evolution prediction—it analyzes the future development direction of the origin node, uncovers vertical evolution nodes with temporal or developmental stage correlations, and ultimately forms vertical evolution node hypotheses. Lifecycle evolution prediction is based on the natural progression of time, analyzing the natural state changes of the consumer group corresponding to the origin node over time, capturing the changes in identity and needs of the group at different life stages and developmental stages, and thus generating corresponding evolution node hypotheses. For example, from the origin node "pregnant mothers," vertical evolution node hypotheses such as "newborn mothers" and "infant mothers" can be predicted. Event-triggered evolution prediction analyzes the impact of specific external sudden or landmark events on the origin node group, identifying changes in group identity, consumption behavior, or needs directly caused by such events, and thus generating corresponding evolution node hypotheses. For example, from the origin node "newcomers to the workforce," combined with marriage events, the vertical evolution node hypothesis of "married working professionals" can be predicted. Trend-driven evolutionary forecasting analyzes changes in external macroeconomic trends such as market policies, industry development, and consumer trends, assesses the impact of these changes on the original node group, and identifies the group transformations driven by these changes. This leads to the generation of corresponding evolutionary node hypotheses. For example, starting with the original node of "traditional car owners," and combining policy incentives and market trends for new energy vehicles, a vertical evolutionary node hypothesis for "new energy vehicle owners" can be predicted. In practical applications, one or more forecasting methods can be combined based on the characteristics of the original node and the actual market situation. A comprehensive analysis then forms a reasonable and data-supported vertical evolutionary node hypothesis.

[0055] Similarly, the system can use the original consumer group node or the standard consumer group node already in the database as the source node. Through at least one of the following analytical methods—behavioral difference analysis, tag combination analysis, and demand analysis—it can uncover sub-groups within the source node that exhibit significant differences in characteristics, behaviors, and needs. This ultimately forms horizontal sub-node hypotheses. Each sub-node is a subset of the source node, possessing the core attributes of the source node while having its own unique characteristics. Specifically, behavioral difference analysis involves collecting and clustering consumer behavior data corresponding to the source node, identifying behavioral differences within the group in terms of consumption frequency, consumption scenarios, product selection, and usage habits. Based on these differences, different sub-groups are divided, generating corresponding sub-node hypotheses. For example, from the source node "working mothers," analyzing consumer behavior can uncover horizontal sub-node hypotheses such as "frequent business trip mothers" and "working-from-home mothers." Tag combination analysis involves multi-dimensional cross-matching and combination analysis of various feature tags from the source node. By combining different tags, it uncovers sub-groups with composite characteristics, generating corresponding sub-node hypotheses. For example, combining the tags "young people" and "pet owners" with relevant source nodes can uncover the horizontal sub-node hypothesis of "young pet owners." Demand analysis involves deeply breaking down the core needs of the original node group, combining user-generated content data to mine differences in functional, emotional, and social needs within the group, and segmenting the group according to different dimensions of needs. This generates corresponding segmentation node hypotheses. For example, starting with the original node "fitness enthusiasts," demand analysis can uncover horizontal segmentation node hypotheses such as "muscle-building enthusiasts" and "fat-loss enthusiasts." In practical applications, one or more analysis methods can be combined to mine segmented groups from different dimensions of the original node, forming clearly defined and well-defined horizontal segmentation node hypotheses, thus achieving refined segmentation of the original node group.

[0056] In summary, both the longitudinal evolution node hypothesis formed through longitudinal evolution prediction and the lateral subdivision node hypothesis formed through lateral subdivision hypothesis will serve as important components of the new node hypothesis. Subsequently, they will be uniformly entered into the structural verification unit to sequentially complete the data stock verification, relation legality verification, and graph consistency verification. New node hypotheses that pass the verification will be included in the candidate list and enter the subsequent task attribute completion and multi-agent adversarial debate stages. Those that fail will be marked as pending observation and will be re-verified after subsequent data supplementation.

[0057] Optionally, after forming a new node hypothesis and before supplementing the task attribute hypothesis for the new node hypothesis, the method further includes: validating the new node hypothesis, wherein the validity verification includes at least one of data stock verification, relation legality verification and graph consistency verification; supplementing the task attribute hypothesis for the new node hypothesis includes: supplementing the task attribute hypothesis for the new node hypothesis that has passed the validity verification.

[0058] It is known that after forming a new node hypothesis and before completing the task attribute hypothesis, the validity verification of the new node hypothesis can screen out new node hypotheses that have data support, legitimate relationships, and logical consistency in advance, avoiding worthless hypotheses from entering the subsequent attribute completion stage, and improving the efficiency and rationality of the overall graph construction. The validity verification includes at least one of data stock verification, relationship legitimacy verification, and graph consistency verification. Only new node hypotheses that pass the verification can enter the task attribute completion stage.

[0059] Specifically, in the actual validity verification process, one or more of the above verification methods can be selected for verification based on the type of new node hypothesis and actual application needs. Only new node hypotheses that pass the selected validity verification step will be included in the candidate list and become the object of subsequent task attribute completion. New node hypotheses that fail the validity verification will be marked by the system as pending observation and will not enter the subsequent steps. The validity verification process can be restarted after more relevant consumer insight data is supplemented.

[0060] Optionally, the new node hypothesis is validated using existing data, including: counting the number of discussions about the new node hypothesis; if the number of discussions exceeds a preset number, the data validation is deemed successful; and / or, the new node hypothesis is validated for relational legitimacy, including: determining the relation type of the new node hypothesis; if the relation type is unique and meets the preset validation criteria corresponding to the relation type, the relational legitimacy validation is deemed successful, wherein the relation type includes evolutionary relations and subdivision relations; and the new node hypothesis is validated for graph consistency, including: determining whether there is a logical contradiction between the new node hypothesis and existing nodes in the knowledge graph; if there is no logical contradiction, the graph consistency validation is deemed successful.

[0061] Specifically, during data inventory verification, quantitative statistical methods can be used to verify whether the new node hypothesis has real market data support, preventing fictitious nodes without corresponding actual consumer groups from entering the subsequent process. The system will first search all user-generated content data related to the new node hypothesis, and count the overall discussion volume of the new node hypothesis across various data sources. The discussion volume includes quantitative indicators such as the number of related content posts, mentions, and interactions. Then, the actual discussion volume obtained from the statistics is compared with the discussion volume threshold preset by the system. When the actual discussion volume of the new node hypothesis is greater than the preset discussion volume threshold, it can be determined that the new node hypothesis has passed the data inventory verification. Data inventory verification can ensure that the consumer group corresponding to the new node hypothesis actually exists and has a certain market attention and research value, rather than an invalid hypothesis caused by data analysis bias.

[0062] Specifically, when verifying the legality of relationships, the system can clearly define the relationship type between the new node and the original node and existing nodes in the knowledge graph, and verify whether the definition of this relationship type is unique and meets preset standards. Relationship types include two categories: evolutionary relationships and subdivision relationships. The system will establish a two-dimensional judgment matrix composed of semantic inclusion and time series, combining the core information such as the feature association and temporal association between the new node and the original node to determine whether the relationship between the new node and the original node is an evolutionary relationship or a subdivision relationship, ensuring that the judgment result of the relationship type is unique and that there is no situation where it belongs to both types of relationships simultaneously. Then, the system will verify whether the relationship type meets the preset standards. If it is determined to be a subdivision relationship, the system will conduct verification according to the preset verification standards for subdivision relationships, checking whether the new node contains all the key attributes of the original node and has its own unique attributes, using a natural language inference model to detect whether the statement "the new node is the original node" is true, and verifying whether the original node frequently appears as a hypernym of the new node in the corpus. If all conditions are met, the subdivision relationship verification passes. If an evolutionary relationship is determined, the system will perform verification according to the preset verification standards for evolutionary relationships. This includes checking whether the rising popularity period of the new node alternates with the declining popularity period of the original node, whether the new node and the original node have high contextual similarity and rarely appear as parallel items, and whether the corpus contains evolution trigger words such as "upgrade" or "replacement." If all conditions are met, the evolutionary relationship verification is successful. When the assumed relationship type of the new node is unique and meets all the preset verification standards for the corresponding relationship type, the new node hypothesis is considered to have passed the relationship validity verification.

[0063] Specifically, during graph consistency verification, the logical consistency of the graph is ensured by checking whether the new node hypothesis, once incorporated into the knowledge graph, will logically contradict existing nodes, relationships, and ontology rules. The system checks for logical contradictions between the new node hypothesis and existing nodes in the knowledge graph from the dimensions of ontology constraints, inference logic, and fact verification. The ontology constraint dimension checks whether the new node hypothesis violates predefined type constraints and whether there are attribute mutual exclusion issues. For example, if the graph defines concepts and entities as mutually exclusive, the new node cannot be classified into both categories simultaneously. The inference logic dimension checks for transitive conflicts. For instance, if the existing graph has a relationship where node A is superior to node B and node B is superior to node C, a new node hypothesis proposing that node C is superior to node A would constitute a loop contradiction. It also checks whether unique attributes in the graph have conflicting values ​​due to the new node hypothesis. The fact verification dimension uses external authoritative knowledge bases or search engines to verify the authenticity of triples related to the new node hypothesis and calculate confidence scores. After the system completes the full-dimensional verification, if no logical contradictions are found and the confidence score of the fact verification reaches the preset standard, it can be determined that the new node hypothesis does not have any logical contradictions with the existing nodes of the knowledge graph, and thus the new node hypothesis passes the graph consistency verification.

[0064] S220, Complete the task attribute assumptions for the new node.

[0065] Among them, the task attribute hypothesis refers to the consumer to-do task idea initially generated for the new node hypothesis by combining the task attribute reference of similar nodes. It is the prototype of the core attribute of the new node hypothesis. It needs to be supported by evidence and verified before it can become the formal task attribute of the node.

[0066] Optionally, the task attribute hypothesis is supplemented for the new node hypothesis, including: parsing the new node hypothesis to obtain the node name, definition and known features of the new node hypothesis, and integrating them into the target population features; retrieving similar nodes in the consumer knowledge graph based on the target population features, and extracting the task attributes of the similar nodes as reference task attributes; and generating task attribute hypotheses through the hypothesis generation agent based on the reference attributes, wherein the task attribute hypotheses include at least one of functional task attribute hypotheses, emotional task attribute hypotheses and social task attribute hypotheses.

[0067] Specifically, the system first extracts existing basic information from the hypothetical new node, including the node name, the definition of the target population, and known characteristics of the population obtained through evolutionary prediction or horizontal segmentation. These known characteristics cover key consumption-related information such as the population's identity attributes, consumption behavior tendencies, and core needs and preferences. The system then systematically integrates the extracted information, removing duplicates and invalid information, and combining the population identifier corresponding to the node name, the population characteristics in the definition, and the specific attributes from the known characteristics to form the target population characteristics. Next, the system uses these integrated target population characteristics as the retrieval basis to perform similarity matching searches in the consumer knowledge graph's population node database. By calculating the similarity between the target population feature vector of the hypothetical new node and the feature vectors of the standard consumer population nodes already in the database, similar nodes with high feature matching degrees are selected. After screening, the complete task attributes of similar nodes that have been verified are extracted, including their functional, emotional, and social task attributes, as well as corresponding confidence levels and supporting evidence. These extracted task attributes are then organized and screened, removing attributes that do not match the characteristics of the target audience of the new node hypothesis, retaining highly relevant attributes, and integrating them into reference task attributes. This provides a direct reference for generating task attributes for the new node hypothesis, ensuring that the newly generated attribute hypotheses conform to the task attribute characteristics of the same group. Finally, the hypothesis generation AI first conducts an in-depth analysis of the reference task attributes to clarify the task attribute setting patterns, attribute type distribution, and attribute expression forms corresponding to core needs of the same group. Then, combined with the characteristics of the target audience of the new node hypothesis, it uncovers the unique characteristics and core needs that distinguish this group from the similar node groups. Subsequently, it generates corresponding task attribute hypotheses from three dimensions: functionality, emotion, and social. Functional task attribute hypotheses focus on the actual usage needs of the target audience in actual consumption and usage scenarios, generating hypotheses around specific product functions and usage effects. Emotional task attribute hypotheses focus on the psychological and emotional needs of the target audience during the consumption process, generating hypotheses around emotional satisfaction, psychological comfort, and emotional experience. The social task attribute hypothesis focuses on the target group's identity and social needs in social scenarios, and generates hypotheses around aspects such as social recognition, group belonging, and social expression.

[0068] S230. The new node hypothesis for completing the task attribute hypothesis is iteratively verified through multi-agent adversarial debate, and the verified new node hypothesis is transformed into standard consumer group nodes and written into the knowledge graph.

[0069] The multi-agent adversarial debate refers to a debate system composed of three types of agents with different stances: supporters, critics, and reviewers. It's a multi-round critical discussion and verification mechanism centered on supplementing the new node hypothesis to complete the task attribute assumptions. Supporters provide evidence and reasoning support for the hypothesis, critics uncover flaws and raise questions, and reviewers neutrally evaluate the quality of the arguments and give a score. Iterative verification refers to the cyclical verification process where, based on the multi-agent adversarial debate, if the new node hypothesis fails verification, an action plan is generated according to the reviewers' opinions, evidence retrieval and task attribute hypothesis revision are carried out again, and the debate phase begins again. A standard consumer group node refers to a standardized consumer group node that conforms to the consumer knowledge graph's database entry specifications after structural verification, task attribute completion, and iterative verification through multi-agent adversarial debate. It includes complete information such as name, definition, coverage, three types of task attributes and evidence, evolutionary tracing, and verification logs. It can be directly written into the knowledge graph as a formal component of the graph or used as a basis for subsequent node expansion.

[0070] Figure 3 This invention provides a method for constructing a consumer knowledge graph, wherein step S130 mainly includes the following steps S231 to S233: S231. The evidence retrieval agent transforms the task attribute hypothesis into search keywords, retrieves consumer insight data based on the search keywords, filters relevant evidence, and generates an evidence report.

[0071] Specifically, the evidence retrieval agent first breaks down the task attribute assumptions of the new node hypothesis, extracting search keywords and statements based on the descriptions of different types of task attribute assumptions, such as functional, emotional, and social ones. Then, using these keywords, it conducts a comprehensive search of massive amounts of consumer insight data, encompassing macro trend data, search trend data, and user-generated content (UGC) data, with UGC data being the core search object. After completing the search, the evidence retrieval agent performs multi-dimensional relevance screening on the search results, eliminating invalid data that is irrelevant to the task attribute assumptions, of low quality, or from a single source, retaining content that highly matches the assumptions, is from multiple sources, and has practical reference value as relevant evidence. Relevant evidence will clearly indicate key information such as source, mention count, and relevant characteristics. Finally, the evidence retrieval agent systematically organizes all the filtered relevant evidence, classifying and categorizing it according to the type of task attribute assumption, and combining it with the basic information of the new node hypothesis to form an evidence report. The report will clearly state the evidence supporting each task attribute assumption, providing detailed and traceable factual evidence for subsequent debate.

[0072] S232. Send the evidence report to the supporter agent and the critic agent respectively for debate, wherein the supporter agent is used to form supporting arguments and the critic agent is used to form challenging arguments.

[0073] Among them, the supporter agent is used to form supporting arguments, and the critic agent is used to form challenging arguments. Through the collision of viewpoints between the two sides, the rationality and potential problems of the new node hypothesis and task attribute hypothesis can be fully explored.

[0074] Specifically, the system will simultaneously send the generated complete evidence report to both the supporter and critic agents. Based on the same evidence report, the two agents will conduct independent analyses from different perspectives and formulate corresponding arguments. The supporter agent, taking the affirmative stance, will use the evidence report as a basis to argue for the rationality, validity, and innovativeness of the new node hypothesis and task attribute hypothesis. Its supporting arguments follow a "claim + evidence + reasoning" logic: first, it clearly states the core claim supporting the new node hypothesis and task attribute hypothesis; then, it extracts relevant evidence from the evidence report as factual support; and finally, through rigorous logical reasoning, it proves the correlation between the evidence and the claim, demonstrating that the new node hypothesis is not a result of accidental data fluctuations, that the task attribute hypothesis highly matches the characteristics of the target population, and that it possesses unique value distinct from existing nodes. The critic agent, taking the opposing stance, also analyzes the evidence report, uncovering flaws and problems in the logic, data, and boundary definition of the new node and task attribute assumptions. Its critical arguments follow a "refutation + counterexample + challenge" logic: first, it specifically refutes any claims the supporting agent might make; then, it looks for counterexamples in the evidence report or identifies flaws in the evidence, such as a single data source, insufficient sample size, or overlap with existing node attributes; finally, based on the counterexamples and evidentiary flaws, it raises targeted challenges, pointing out potential irrationality in the new node and task attribute assumptions, such as aliases for existing nodes or a lack of actual market support. After both agents complete their arguments, they synchronize their supporting and challenging arguments to the reviewer agent.

[0075] S233. The reviewer agent evaluates the supporting and challenging arguments, generates a quality score, and performs iterative verification based on the quality score.

[0076] Among them, the neutral reviewer AI will evaluate the content of the debate, determine whether the new node hypothesis is verified by quantitative scoring, and start the corresponding iteration process based on the scoring results.

[0077] Specifically, the reviewer agent first analyzes the supporting arguments of the supporter agent and the questioning arguments of the critic agent. Combined with the evidence report, it conducts a comprehensive evaluation from four dimensions: sufficiency of evidence, logical consistency, innovation and differentiation, and antifragility. Each dimension has clear scoring rules and weighting percentages, with sufficiency of evidence and logical consistency each accounting for 30%, and innovation and differentiation and antifragility each accounting for 20%. The reviewer agent will score each dimension according to the rules and then calculate the comprehensive quality score based on the weights. The maximum quality score is 10 points, and the system's preset passing threshold is 8 points.

[0078] Optionally, iterative verification is performed based on the quality score, including: obtaining a preset score threshold and a preset debate round, and determining the current debate round; when the quality score reaches the preset score threshold for two consecutive rounds, the new node hypothesis is confirmed to have passed verification; when the quality score does not reach the preset score threshold for two consecutive rounds and the current debate round does not exceed the preset debate round, the debate focus is determined by the evidence retrieval agent, relevant evidence is supplemented based on the debate focus, and the evidence report is updated. The updated evidence report is sent to the supporter agent and the critic agent for a re-debate; when the quality score does not reach the preset score threshold for two consecutive rounds and the current debate round reaches the preset debate round, the cumulative score is calculated, and a final decision is made based on the cumulative score.

[0079] The preset number of debate rounds represents the maximum number of debates allowed by the system, which can be up to three. Once the preset number of debate rounds is exceeded, no new debates will be held; instead, the final decision will be based on cumulative scores. Simultaneously, the system will record the progress of the adversarial debates in real time, clearly indicating the current debate round. For example, after the first debate, the current round is round 1; after the second round, the round updates to round 2. This information is used to determine whether conditions are suitable for continuing the debate.

[0080] Specifically, the reviewer AI will generate a corresponding quality score after each round of debate. The system will continuously track and compare the scores of two consecutive rounds. If the score in the first round of debate reaches 8 points or above, and the quality score in the second round of debate also reaches 8 points or above, meeting the criteria for two consecutive rounds, the system will directly determine that the new node hypothesis and its corresponding task attribute hypothesis have passed the multi-agent adversarial debate verification. Through two consecutive rounds of high-score verification, it is confirmed that the new node hypothesis has sufficient evidence support, rigorous logical consistency, significant innovative differences, and good antifragility, and can fully address the opposing side's questions. There are no obvious loopholes or problems, and no further debate is needed. It can directly enter the report synthesis stage, where the report synthesis AI will transform it into a standard consumer group node.

[0081] Furthermore, if the quality score for two consecutive rounds fails to reach the preset threshold of 8 points (e.g., 6 points in one round, 7 points in another, or 7 points in both rounds), and the system records that the current number of debate rounds has not exceeded the preset maximum of 3 rounds (e.g., only 1 or 2 rounds of debate have been completed), the system will initiate an evidence supplementation and re-debate process. First, the evidence retrieval agent, combining the reviewer agent's scoring opinions and the debate content from supporters and critics, accurately extracts the debate focus of this round. The debate focus refers to the key issues that raise questions about the new node hypothesis, such as a single source of evidence, insufficient sample size, blurred boundaries with existing nodes, or flaws in logical reasoning. Then, the evidence retrieval agent transforms the debate focus into new search keywords and statements, expanding the search scope to conduct in-depth searches of consumer insight data, focusing on finding multi-source, highly relevant supplementary evidence for the debate focus issues, and eliminating invalid or low-quality content from the original evidence. Finally, the evidence retrieval agent systematically organizes the supplemented evidence, combining it with the structure of the original evidence report to create a completely new evidence report. After the evidence report is updated, the system will send the new evidence report to both the supporter and critic agents. Both parties will then conduct a new round of adversarial debate based on the updated evidence. The supporter agent can combine supplementary evidence to improve its supporting arguments and respond to questions, while the critic agent can continue to raise targeted questions based on the new evidence. The reviewer agent will then evaluate the content of the new round of debate and generate a new quality score, which will then proceed to the next round of iterative judgment.

[0082] Ultimately, if the quality score fails to reach the preset threshold of 8 points after two consecutive rounds, and the system has recorded the maximum preset number of debate rounds (3 rounds), the system will stop initiating new debates and begin the cumulative score statistics and final adjudication process. The reviewer's AI will statistically analyze the quality score generated in each of the three rounds of debate, calculate the cumulative score across all three rounds, and combine this with the dimensional details of each round's score to generate a comprehensive score analysis report. Then, the system will make a final judgment based on the preset cumulative score adjudication criteria and the cumulative score. If the cumulative score reaches the preset passing score, the new node hypothesis is considered validated and can proceed to the subsequent report synthesis stage. If the cumulative score does not reach the passing score, the new node hypothesis is considered unvalidated, and the system will mark the hypothesis as needing improvement, temporarily excluding it from the consumer knowledge graph. Simultaneously, the system will record the core issues and review comments to provide a reference for subsequent optimization and re-validation.

[0083] The technical solution of this invention, through vertical evolution prediction and / or horizontal segmentation assumptions formed by the creation consumer group node or existing standard consumer group nodes, enables the bidirectional expansion of the population node, capturing the life cycle changes and emerging segmentation needs of the consumer group, and breaking through the limitations of the fixed traditional segmentation standards. By supplementing the task attribute assumptions for the new node assumptions, the core attributes of the nodes can be improved, the deep motivations of consumers can be explored, and the problem of the lack of modeling for non-behavioral needs in existing technologies can be solved. Through multi-agent adversarial debate, the new node assumptions are iteratively verified, and the verified nodes are written into the graph, which can ensure the credibility and evidence support of the node information, while realizing the continuous automatic evolution of the graph, so that knowledge updates keep up with real-time market changes and reduce manual maintenance costs.

[0084] Example 3 Figure 4 This is a schematic diagram of a consumer knowledge graph construction device provided in Embodiment 3 of the present invention. Figure 4 As shown, the device includes: a consumer insight data acquisition module 310, used to acquire consumer insight data from a target data source, the target data source including at least one; The to-do task acquisition module 320 is used to acquire consumers' to-do tasks based on consumer insight data; Genesis Node Generation Module 330 is used to generate Genesis Consumer Group Nodes corresponding to Consumer To-Do Tasks based on Consumer To-Do Tasks. The Consumer Knowledge Graph Construction Module 340 is used to construct a consumer knowledge graph based on the genesis consumer group nodes.

[0085] Optionally, the task acquisition module 320 is specifically used for: identifying target trend signals in consumer insight data that meet predetermined change trends based on consumer insight data; determining abnormal phenomena that differ from historical phenomena based on target trend signals; and performing cluster analysis on consumer insight data related to abnormal phenomena to identify emerging consumer tasks.

[0086] Optionally, the Genesis Node Generation Module 330 is specifically used to: define the name, definition, coverage, and characteristics of the emerging consumer group corresponding to the emerging consumer to-do tasks; and generate a structured Genesis Consumer Group Node based on the name, definition, coverage, and characteristics of the emerging consumer group.

[0087] Optionally, the device also includes: a graph dynamic evolution module, used to: perform vertical evolution prediction and / or horizontal subdivision hypothesis on the generative consumer group node or the standard consumer group node already in the database to form new node hypothesis; complete the task attribute hypothesis for the new node hypothesis; iteratively verify the new node hypothesis that completes the task attribute hypothesis through multi-agent adversarial debate, and transform the verified new node hypothesis into a standard consumer group node and write it into the knowledge graph.

[0088] Optionally, the dynamic evolution module of the graph specifically includes: a new node hypothesis forming unit, used to: take the generative consumer group node or the standard consumer group node already in the database as the original node, and perform at least one of the following predictions on the original node: life cycle evolution prediction, event-triggered evolution prediction, and trend-driven evolution prediction, to form a vertical evolution node hypothesis; and / or, to perform horizontal segmentation hypothesis on the generative consumer group node or the standard consumer group node already in the database, to form a new node hypothesis, including: taking the generative consumer group node or the standard consumer group node already in the database as the original node, and performing at least one of the following analyses on the original node: behavioral difference analysis, tag combination analysis, and demand analysis, to form a horizontal segmentation node hypothesis.

[0089] Optionally, the graph dynamic evolution module further includes: a new node hypothesis verification unit, used to: verify the validity of new node hypotheses, wherein the validity verification includes at least one of data stock verification, relation legality verification and graph consistency verification; and to complete task attribute hypotheses for new node hypotheses, including: completing task attribute hypotheses for new node hypotheses that have passed the validity verification.

[0090] Optionally, the new node hypothesis verification unit is specifically used for: counting the discussion volume of the new node hypothesis; when the discussion volume is greater than the preset discussion volume, the data stock verification is confirmed to be passed; and / or, performing relational legality verification on the new node hypothesis, including: determining the relation type of the new node hypothesis; when the relation type is unique and meets the preset verification criteria corresponding to the relation type, the relational legality verification is confirmed to be passed, wherein the relation type includes evolutionary relations and subdivision relations; and performing graph consistency verification on the new node hypothesis, including: determining whether there is a logical contradiction between the new node hypothesis and the existing nodes of the knowledge graph; if there is no logical contradiction, the graph consistency verification is confirmed to be passed.

[0091] Optionally, the graph dynamic evolution module specifically includes: a task hypothesis completion unit, used to: parse new node hypotheses to obtain the node name, definition, and known features of the new node hypotheses, and integrate them into target population features; retrieve similar nodes in the consumer knowledge graph based on the target population features, and extract the task attributes of similar nodes as reference task attributes; and generate task attribute hypotheses through a hypothesis generation agent based on the reference attributes, wherein the task attribute hypotheses include at least one of functional task attribute hypotheses, emotional task attribute hypotheses, and social task attribute hypotheses.

[0092] Optionally, the graph dynamic evolution module specifically includes: an agent debate unit, used to: transform task attribute hypotheses into search keywords through an evidence retrieval agent, retrieve consumer insight data based on the search keywords, filter relevant evidence, and generate an evidence report; send the evidence report to supporter agents and critic agents respectively for debate, where supporter agents are used to form supporting arguments and critic agents are used to form challenging arguments; and reviewer agents evaluate the supporting and challenging arguments, generate quality scores, and perform iterative verification based on the quality scores.

[0093] Optionally, the agent debate unit specifically includes: an iterative verification subunit, used to: obtain a preset scoring threshold and a preset debate round, and determine the current debate round; when the quality scores for two consecutive rounds reach the preset scoring threshold, determine that the new node hypothesis has been verified; when the quality scores for two consecutive rounds do not reach the preset scoring threshold and the current debate round does not exceed the preset debate round, determine the debate focus through the evidence retrieval agent, supplement relevant evidence based on the debate focus and update the evidence report, and send the updated evidence report to the supporter agent and the critic agent to re-debate; when the quality scores for two consecutive rounds do not reach the preset scoring threshold and the current debate round reaches the preset debate round, calculate the cumulative score, and make a final decision based on the cumulative score.

[0094] The technical solution of this invention, by acquiring multi-source consumer insight data, can comprehensively capture market consumption information, providing a rich data foundation for subsequent analysis. By acquiring consumers' to-do tasks, it can accurately uncover consumers' deep needs, breaking through the limitations of traditional methods that only focus on behavioral data. It can automatically discover emerging consumer groups, solving the problem that manual definition is insufficient for uncovering new areas. By constructing a consumer knowledge graph, a structured network of connections centered on consumer needs can be built, providing precise demographic knowledge support for enterprise decision-making and realizing the graph-based implementation from needs to demographics.

[0095] The consumer knowledge graph construction apparatus provided in this embodiment of the invention can execute the consumer knowledge graph construction method provided in any embodiment of the invention, and has the corresponding functional modules and beneficial effects of the execution method.

[0096] Example 4 Figure 5 A schematic diagram of an electronic device 10 that can be used to implement embodiments of the present invention is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital assistants, cellular phones, smartphones, wearable devices (e.g., helmets, glasses, watches, etc.), and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the invention described and / or claimed herein.

[0097] like Figure 5 As shown, the electronic device 10 includes at least one processor 11 and a memory, such as a read-only memory (ROM) 12 or a random access memory (RAM) 13, communicatively connected to the at least one processor 11. The memory stores computer programs executable by the at least one processor. The processor 11 can perform various appropriate actions and processes based on the computer program stored in the ROM 12 or loaded from storage unit 18 into the RAM 13. The RAM 13 can also store various programs and data required for the operation of the electronic device 10. The processor 11, ROM 12, and RAM 13 are interconnected via a bus 14. An input / output (I / O) interface 15 is also connected to the bus 14.

[0098] Multiple components in electronic device 10 are connected to I / O interface 15, including: input unit 16, such as keyboard, mouse, etc.; output unit 17, such as various types of displays, speakers, etc.; storage unit 18, such as disk, optical disk, etc.; and communication unit 19, such as network card, modem, wireless transceiver, etc. Communication unit 19 allows electronic device 10 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0099] Processor 11 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of processor 11 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. Processor 11 performs the various methods and processes described above, such as a method for constructing a consumer knowledge graph.

[0100] In some embodiments, a method for constructing a consumer knowledge graph can be implemented as a computer program tangibly contained in a computer-readable storage medium, such as storage unit 18. In some embodiments, part or all of the computer program can be loaded and / or installed on electronic device 10 via ROM 12 and / or communication unit 19. When the computer program is loaded into RAM 13 and executed by processor 11, one or more steps of the method for constructing a consumer knowledge graph described above can be performed. Alternatively, in other embodiments, processor 11 can be configured to execute a method for constructing a consumer knowledge graph by any other suitable means (e.g., by means of firmware).

[0101] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-a-chip (SoCs), payload-programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0102] Computer programs used to implement the methods of the present invention may be written in any combination of one or more programming languages. These computer programs may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, such that when executed by the processor, the computer programs cause the functions / operations specified in the flowcharts and / or block diagrams to be performed. The computer programs may be executed entirely on a machine, partially on a machine, or as a standalone software package, partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0103] In the context of this invention, a computer-readable storage medium can be a tangible medium that may contain or store a computer program for use by or in conjunction with an instruction execution system, apparatus, or device. A computer-readable storage medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination thereof. Alternatively, a computer-readable storage medium may be a machine-readable signal medium. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof.

[0104] To provide interaction with a user, the systems and techniques described herein can be implemented on an electronic device having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the electronic device. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0105] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as data servers), or middleware components (e.g., application servers), or frontend components (e.g., user computers with graphical user interfaces or web browsers through which users can interact with implementations of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., communication networks). Examples of communication networks include local area networks (LANs), wide area networks (WANs), blockchain networks, and the Internet.

[0106] A computing system can include clients and servers. Clients and servers are generally located far apart and typically interact through communication networks. The client-server relationship is created by computer programs running on the respective computers and having a client-server relationship with each other. The server can be a cloud server, also known as a cloud computing server or cloud host, which is a hosting product within the cloud computing service system. It addresses the shortcomings of traditional physical hosts and VPS services, such as high management difficulty and weak business scalability.

[0107] It should be understood that the various forms of processes shown above can be used, with steps reordered, added, or deleted. For example, the steps described in this invention can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution of this invention can be achieved, and this is not limited herein.

[0108] The specific embodiments described above do not constitute a limitation on the scope of protection of this invention. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this invention should be included within the scope of protection of this invention.

Claims

1. A method for constructing a consumer knowledge graph, characterized in that, include: Acquire consumer insight data from a target data source, wherein the target data source includes at least one; Based on the aforementioned consumer insight data, consumer to-do tasks are obtained; Based on the consumer to-do tasks, a genealogical consumer group node corresponding to the consumer to-do tasks is generated. Based on the aforementioned genesis consumer group nodes, a consumer knowledge graph is constructed.

2. The method according to claim 1, characterized in that, The consumer insight data includes at least one of macro trend data, search trend data, and user-generated content (UGC) data. The process of obtaining consumer to-do tasks based on the aforementioned consumer insight data includes: Based on the consumer insight data, identify target trend signals in the consumer insight data that meet a predetermined trend of change; Based on the target trend signal, identify abnormal phenomena that differ from historical phenomena; Cluster analysis was performed on the consumer insight data related to the aforementioned anomalies to identify emerging consumer to-do tasks.

3. The method according to claim 2, characterized in that, The predetermined trend of change includes at least one of the following: The number of mentions in a predetermined time window increases by more than a first specified threshold compared to the number of mentions in the previous time window. It exhibits a monotonically increasing trend within a consecutive first specified number of predetermined time periods; From a single data source, the reference spreads to at least a second specified number of data sources; The first appearance of the term or a change from a predetermined low-frequency range mention to a predetermined high-frequency range mention.

4. The method according to claim 2, characterized in that, The process of generating the corresponding Genesis Consumer group node based on the consumer's to-do task includes: Based on the emerging consumer to-do tasks, the name, definition, coverage and characteristics of the emerging consumer group corresponding to the emerging consumer to-do tasks are defined. Based on the name, definition, coverage, and characteristics of the emerging consumer group, a structured genealogy consumer group node is generated.

5. The method according to any one of claims 1 to 4, characterized in that, After constructing the consumer knowledge graph, the method further includes: For the aforementioned generative consumer group nodes or standard consumer group nodes already in the database, conduct vertical evolution prediction and / or horizontal segmentation assumptions to form new node hypotheses; To complete the task attribute assumptions for the new node; The new node hypothesis of the completion task attribute assumption is iteratively verified by multi-agent adversarial debate, and the verified new node hypothesis is transformed into standard consumer group nodes and written into the knowledge graph.

6. The method according to claim 5, characterized in that, The step of conducting longitudinal evolution prediction on the generative consumer group node or the standard consumer group node already included in the database to form a new node hypothesis includes: taking the generative consumer group node or the standard consumer group node already included in the database as the original node, and performing at least one of the following predictions on the original node: life cycle evolution prediction, event-triggered evolution prediction, and trend-driven evolution prediction, in order to form a longitudinal evolution node hypothesis. And / or, The process of making horizontal segmentation assumptions on the generative consumer group nodes or the standard consumer group nodes already in the database to form new node assumptions includes: Using the original consumer group node or the standard consumer group node already in the database as the original node, perform at least one of the following analyses on the original node: behavioral difference analysis, tag combination analysis, and demand analysis, to form the horizontal segmentation node hypothesis.

7. The method according to claim 5, characterized in that, After forming the new node hypothesis and before completing the task attribute hypothesis for the new node hypothesis, the method further includes: The new node hypothesis is validated, wherein the validation includes at least one of data inventory validation, relation legitimacy validation, and graph consistency validation. The assumptions for completing the task attributes of the new node include: Complete the task attribute assumptions for the new node assumptions that have passed the validity verification.

8. The method according to claim 7, characterized in that, The data storage verification of the new node hypothesis includes: counting the discussion volume of the new node hypothesis, and determining that the data storage verification is passed when the discussion volume is greater than a preset discussion volume; And / or, The step of verifying the legality of the new node hypothesis includes: determining the relationship type of the new node hypothesis; when the relationship type is unique and meets the preset verification criteria corresponding to the relationship type, the legality verification of the relationship is determined to be passed, wherein the relationship type includes evolutionary relationship and subdivision relationship; The graph consistency verification of the new node hypothesis includes: Determine whether there is a logical contradiction between the new node hypothesis and the existing nodes of the knowledge graph. If there is no logical contradiction, the graph consistency verification is passed.

9. The method according to claim 5, characterized in that, The above is a hypothesis supplement for the new node. The full set of task attribute assumptions includes: The new node hypothesis is analyzed to obtain the node name, definition, and known features of the new node hypothesis, and then integrated into the target population features; Based on the characteristics of the target audience, similar nodes in the consumer knowledge graph are retrieved, and the task attributes of the similar nodes are extracted as reference task attributes. Based on the reference attributes, a hypothesis-generating agent generates task attribute hypotheses, wherein the task attribute hypotheses include at least one of functional task attribute hypotheses, emotional task attribute hypotheses, and social task attribute hypotheses.

10. The method according to claim 9, characterized in that, The iterative verification of the new node hypothesis of the completion task attribute hypothesis through multi-agent adversarial debate includes: The evidence retrieval agent transforms the task attribute hypothesis into search keywords, retrieves the consumer insight data based on the search keywords, filters relevant evidence, and generates an evidence report. The evidence reports are sent to supporter agents and critic agents respectively for debate, whereby the supporter agents are used to form supporting arguments and the critic agents are used to form challenging arguments. The supporting arguments and the challenging arguments are evaluated by a reviewer agent to generate a quality score, and iterative verification is performed based on the quality score.

11. The method according to claim 10, characterized in that, The iterative verification based on the quality score includes: Obtain the preset scoring threshold and preset debate rounds, and determine the current debate round; When the quality scores reach the preset scoring threshold for two consecutive rounds, it is determined that the hypothesis of the new node has been verified. When the quality score fails to reach the preset score threshold for two consecutive rounds and the current debate round does not exceed the preset debate round, the evidence retrieval agent determines the debate focus, supplements relevant evidence based on the debate focus, updates the evidence report, and sends the updated evidence report to the supporter agent and critic agent to conduct a new debate. If the quality score fails to reach the preset score threshold for two consecutive rounds and the current debate round reaches the preset debate round, the cumulative score is calculated, and a final decision is made based on the cumulative score.

12. A device for constructing a consumer knowledge graph, characterized in that, include: The consumer insight data acquisition module is used to acquire consumer insight data from target data sources, including at least one target data source. The task acquisition module is used to acquire consumers' tasks based on consumer insight data. The Genesis Node Generation Module is used to generate Genesis Consumer Group Nodes corresponding to Consumer To-Do Tasks based on Consumer To-Do Tasks. The Consumer Knowledge Graph Construction Module is used to build a consumer knowledge graph based on the genesis consumer group nodes.

13. An electronic device, characterized in that, The electronic device includes: At least one processor; and A memory communicatively connected to the at least one processor; wherein, The memory stores a computer program that can be executed by the at least one processor, the computer program being executed by the at least one processor to enable the at least one processor to perform the method for constructing a consumer knowledge graph according to any one of claims 1-11 of the present invention.

14. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions that cause a processor to execute the method for constructing a consumer knowledge graph according to any one of claims 1-11.

Images