Personalized call handling methods and systems

Abstract

This invention discloses a personalized call processing method and system, relating to the field of artificial intelligence technology. The method includes: parsing the call intent of an unknown call received by a terminal; extracting the user preference profile embedding vector and the call intent embedding vector of the current user from a pre-trained dynamic heterogeneous graph; calculating the matching score of the user preference profile embedding vector and the call intent embedding vector; and generating a decision instruction based on the matching score. This helps to solve the problem that existing technologies cannot achieve personalized call processing for unknown numbers.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and in particular to a personalized call processing method and system. Background Technology

[0002] In the field of mobile communication technology, blocking harassing calls from unknown numbers is a crucial requirement for ensuring a good user communication experience. Current mainstream solutions fall into two main categories: one is rule-based blocking based on a number database, where operators, security vendors, and users jointly maintain a blacklist of numbers to block already marked harassing calls. However, this method suffers from a significant lag in detecting newly emerging harassing numbers. The other is classification-based blocking based on user-generated labels, allowing users to manually label incoming calls as "harassing calls," "real estate agents," or other pre-defined, coarse categories after receiving them. Once a certain number of labels are collected, the system creates a "profile" of the number and alerts or blocks calls from other users.

[0003] However, existing technologies have significant drawbacks: First, the "one-size-fits-all" approach lacks personalization, and the universal blocking mechanism cannot distinguish individual user needs, easily leading to false or missed blocking; second, the labeling dimensions are coarse, and the limited fixed labels make it difficult to accurately identify the true intent of incoming calls and distinguish between calls of different values ​​under the same label; third, the user feedback mechanism is singular and biased, relying excessively on sparse and lagging active labeling, ignoring implicit behaviors that contain true preferences, such as immediate hanging up, call duration, and voice emotion; fourth, the model is static, and the simple statistical static label set cannot uncover deep correlations of preferences, nor can it perceive the dynamic changes of preferences over time.

[0004] Therefore, there is an urgent need for a method that can personalize caller ID processing for unknown numbers. Summary of the Invention

[0005] In view of this, the present invention proposes a personalized call processing method and system that can realize personalized call processing for unknown numbers.

[0006] To achieve the above objectives, the present invention provides the following technical solution:

[0007] A personalized call handling method includes:

[0008] Analyze the call intent of unknown calls received by the terminal;

[0009] Extract the user preference profile embedding vector and caller intent embedding vector of the current user from a pre-trained dynamic heterogeneous graph;

[0010] Calculate the matching score between the user preference profile embedding vector and the caller intent embedding vector;

[0011] Decision instructions are generated based on the matching scores.

[0012] Based on the above technical solution, the present invention can be further improved as follows:

[0013] Optionally, the call intent of the unknown caller received by the parsing terminal includes:

[0014] After receiving an unknown call from a terminal, the cloud server interacts with the unknown caller to obtain interaction data.

[0015] The cloud server parses the interaction data to obtain the call intent of the unknown caller.

[0016] Optionally, before the steps of extracting the current user's user preference profile embedding vector and caller intent embedding vector from the pre-trained dynamic heterogeneous graph, the following steps are included:

[0017] Construct a node set for a dynamic heterogeneous graph, the node set including user nodes, intent nodes, and entity nodes:

[0018] Create association edges for each node in the dynamic heterogeneous graph, including user-intent edges, intent-entity edges, and intent-intent edges.

[0019] Optionally, before the step of extracting the current user's user preference profile embedding vector and caller intent embedding vector from the pre-trained dynamic heterogeneous graph, the method further includes:

[0020] The dynamic heterogeneous graph is input into the heterogeneous graph neural network model to obtain the embedding vector of each node in the dynamic heterogeneous graph;

[0021] A user preference profile is constructed based on the embedding vector of each node.

[0022] Optionally, the personalized call processing method further includes:

[0023] After the call ends, collect the user's feedback data from this interaction;

[0024] The heterogeneous graph neural network model is updated and trained based on the feedback data.

[0025] Optionally, calculating the matching score between the user preference profile embedding vector and the caller intent embedding vector includes:

[0026] The matching score is calculated using formula (1);

[0027] Formula (1);

[0028] In the formula, To match the score, Embed numerical values ​​for each dimension of the vector to create a user preference profile. The numerical values ​​of each dimension of the vector are embedded to represent the caller's intent.

[0029] Optionally, the step of generating a decision instruction based on the matching score includes:

[0030] If the matching score is less than the first preset threshold, a decision instruction to directly intercept is generated;

[0031] If the matching score is greater than the first preset threshold and less than the second preset threshold, a decision instruction to substitute the connection is generated.

[0032] If the matching score is greater than the second preset threshold, a decision instruction for transferring the user is generated.

[0033] A personalized call handling system includes:

[0034] The intent parsing module is used to parse the call intent of unknown callers;

[0035] The vector extraction module is used to extract the user preference profile embedding vector and the caller intent embedding vector of the current user from a pre-trained dynamic heterogeneous graph.

[0036] The matching calculation module is used to calculate the matching score between the user preference profile embedding vector and the caller intent embedding vector;

[0037] The decision generation module is used to generate decision instructions based on the matching score.

[0038] An electronic device includes a memory, a processor, and a computer program stored in the memory and running on the processor, wherein the processor executes the computer program to implement the steps of the method described herein.

[0039] A non-transitory computer-readable storage medium having a computer program stored thereon, the computer program implementing the steps of the method when executed by a processor.

[0040] The present invention has the following advantages:

[0041] The personalized call handling method in this invention generates decisions by extracting user preference profile embedding vectors and call intent embedding vectors, achieving personalized interception and transfer for each user. This avoids mistakenly blocking valid calls from users with needs or missing harassing calls from users without needs. The dynamic heterogeneous graph can capture the correlation and changes in user preferences, which is more in line with users' real-time needs than static models. At the same time, it does not rely too much on users' active marking, reducing feedback lag and ultimately significantly improving the user's communication experience and call management efficiency. Attached Figure Description

[0042] For illustrative and not limiting purposes, the present invention will now be described in conjunction with embodiments and accompanying drawings, wherein:

[0043] Figure 1 This is a flowchart illustrating the personalized call handling method in an embodiment of the present invention.

[0044] Figure 2 A schematic diagram showing the main components of the personalized call processing system in an embodiment of the present invention;

[0045] Figure 3 This is a schematic diagram of the physical structure of the electronic device provided by the present invention. Detailed Implementation

[0046] To enable those skilled in the art to better understand the present invention, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. 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.

[0047] It should be noted that the terms "first," "second," etc., in the specification 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 used interchangeably where appropriate for the embodiments of the invention described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover 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.

[0048] It should be noted that, where there is no conflict, the embodiments and features of the present invention can be combined with each other. The embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0049] Figure 1 This is a flowchart illustrating the personalized call processing method according to an embodiment of the present invention, as shown below. Figure 1 As shown, the personalized call processing method provided in this embodiment of the invention includes the following steps S101 to S104.

[0050] S101, analyze the call intent of an unknown caller received by the terminal.

[0051] Terminal equipment: responsible for receiving incoming calls, executing server instructions, and collecting raw user behavior data (such as answering / rejecting calls, call duration, user voice stream, etc.).

[0052] Cloud server: Responsible for core computing, mainly including the following modules:

[0053] AI Call Assistant Module: Responsible for engaging in natural language dialogue with the caller and generating call text.

[0054] Call Intent Analysis Module: Receives call text and uses an NLP model to extract structured and refined call intent. Example call intent: {Domain: Real Estate, Business: Shop Promotion, Region: Beijing Chaoyang District, Organization: Lianjia Real Estate}.

[0055] After receiving an unknown call, the terminal reports the unknown call to the cloud server for interaction and obtains interaction data.

[0056] The cloud server parses the interaction data to obtain the call intent of the unknown caller.

[0057] S102, extract the user preference profile embedding vector and caller intent embedding vector of the current user from the pre-trained dynamic heterogeneous graph.

[0058] Before the steps of extracting the current user's preference profile embedding vector and caller intent embedding vector from the pre-trained dynamic heterogeneous graph, the following steps are included:

[0059] Construct a node set for a dynamic heterogeneous graph, the node set including user nodes, intent nodes, and entity nodes:

[0060] User node: Each user corresponds to one node.

[0061] Intent Node: Each structured intent corresponds to one node.

[0062] Entity node: A node corresponds to a key entity extracted from the intent (such as "Lianjia Real Estate" or "China Merchants Bank").

[0063] Create association edges for each node in the dynamic heterogeneous graph, including user-intent edges, intent-entity edges, and intent-intent edges.

[0064] User-Intent Edge: A connection is established when a user interacts with a specific intent. The edge weight is a multi-dimensional feature vector dynamically generated by an attention fusion network based on the intent type, incorporating: ① explicit feedback (user's "like / dislike" rating), ② implicit decision ("transfer to the user / hang up" behavior), ③ call duration (normalized value), ④ call sentiment (user's voice sentiment analysis score), and ⑤ interaction timestamp.

[0065] Intent-Entity Edge: Connects an intent node to its contained entity nodes.

[0066] Intent-Intent Edge: A connection is established based on the semantic similarity of the intent description (e.g., calculated via BERT) for preference generalization.

[0067] The dynamic heterogeneous graph is input into the heterogeneous graph neural network model to obtain the embedding vector of each node in the dynamic heterogeneous graph;

[0068] A user preference profile is constructed based on the embedding vector of each node.

[0069] S103, calculate the matching score between the user preference profile embedding vector and the caller intent embedding vector.

[0070] The matching score is calculated using formula (1);

[0071]

[0072] Formula (1);

[0073] In the formula, To match the score, Embed numerical values ​​for each dimension of the vector to create a user preference profile. The numerical values ​​of each dimension of the vector are embedded to represent the caller's intent.

[0074] S104, Generate decision instructions based on matching scores.

[0075] If the matching score is less than the first preset threshold, a decision instruction is generated to directly block the current incoming call;

[0076] If the matching score is greater than the first preset threshold and less than the second preset threshold, a decision instruction to substitute the connection is generated.

[0077] If the matching score is greater than the second preset threshold, a decision instruction is generated to transfer the current call to the user.

[0078] The personalized call handling method further includes:

[0079] After the call ends, collect the user's feedback data from this interaction;

[0080] The heterogeneous graph neural network model is updated and trained based on the feedback data.

[0081] Heterogeneous Graph Neural Networks (HAN or RGCN) are employed for end-to-end learning of dynamic heterogeneous graphs. The GNN learns a low-dimensional, dense embedding vector for each node in the graph (especially user nodes) by aggregating neighbor node information. Ultimately, the embedding vector `Embedding_User` for each user node constitutes a personalized preference profile for that user, containing deep relationships and dynamic temporal information. The model is incrementally updated periodically (e.g., daily) or when new interactions occur.

[0082] This invention goes beyond simple label matching, using Generative Neural Networks (GNNs) to uncover higher-order relationships behind user preferences. For example, the system can discover through graph structures that "users who like low-interest loan products from Bank A are usually also interested in health insurance from Insurance Company B," thereby achieving intelligent recommendation and interception based on collaborative filtering—something traditional methods cannot achieve.

[0083] For a new merchant or business that a user has never encountered before, as long as the intent node or entity node of the business is associated with a node that the user already has preferences in the graph (such as semantic similarity or belonging to the same domain), the present invention can effectively predict the user's potential preferences, greatly improving the efficiency of identifying and blocking new harassment patterns.

[0084] This invention introduces timestamp features into the interaction edges and combines them with the dynamic update mechanism of GNN, making the user profile "alive." It can perceive the generation, enhancement, and decay of user interests, enabling the interception strategy to automatically adapt to changes in user needs throughout different lifecycles.

[0085] For new users or users with limited interaction, the GNN can aggregate a large amount of other users' preference information through a few connected intent nodes, thereby quickly generating a high-quality initial profile and greatly alleviating the cold start problem.

[0086] By introducing an attention mechanism, this invention can intelligently determine the importance of different user feedback (explicit, implicit, duration, etc.) based on specific circumstances (such as intent type), making the fused interaction information more accurate and representative, thereby improving the accuracy of the profile.

[0087] Alternative solution to the present invention:

[0088] The system maintains two simple sets for each user: a "preference tag set" and a "dislike tag set". Users can explicitly indicate "like" or "dislike" the intent of the AI ​​assistant after it answers, adding the corresponding intent tag (such as "loan sales") to these two sets respectively.

[0089] The AI ​​call assistant parses the call content into a single, standardized intent label, such as "loan sales." It then performs a strict string equality check. If the call intent label exists in the user's "disliked label set," it executes "direct blocking." If the call intent label exists in the user's "liked label set," it executes "direct transfer." If the label does not appear in either set, the default strategy is executed (e.g., the AI ​​assistant continues answering).

[0090] Alternatively, user preferences can be upgraded from a simple set to a weighted "tag-preference score" dictionary. For example: {"real estate agent": -1.0, "delivery and takeout": 0.8}. Structured intent can be broken down into a set of keyword tags, such as ["finance", "loan promotion", "China Merchants Bank"]. The overall user preference can be determined by calculating the sum of the scores of tags in the user profile that overlap with the incoming call keywords.

[0091] Alternative solution to the present invention:

[0092] Instead of using graph neural networks, a "label-preference" dictionary or multi-hot vector is maintained for each user. For example, user U1's profile could be {"real estate agent": -1.0,"delivery and takeout": 0.8,"bank follow-up": 0.5,...}, where positive values ​​represent liking, negative values ​​represent disliking, and the absolute value represents the intensity.

[0093] The structured intents analyzed by the AI ​​assistant are simply mapped to a set of keyword tags. For example, the intent {domain: real estate, business: shop promotion, ...} is represented as ["real estate", "shop promotion"].

[0094] Decisions are made by calculating the weighted intersection score of user profile tags and caller intent tags. For example, if the caller intent tag "real estate" has a negative score in the user profile, the caller is more likely to be blocked.

[0095] Figure 2 This is a schematic diagram illustrating the main components of the personalized call processing system according to an embodiment of the present invention. (See diagram below.) Figure 2 As shown, the personalized call processing system 1 provided in this embodiment of the invention includes an intent parsing module 10, a vector extraction module 20, a matching calculation module 30, and a decision generation module 40.

[0096] The intent parsing module 10 is used to parse the call intent of an unknown call received by the terminal;

[0097] Vector extraction module 20 is used to extract the user preference profile embedding vector and caller intent embedding vector of the current user from the pre-trained dynamic heterogeneous graph;

[0098] The matching calculation module 30 is used to calculate the matching score between the user preference profile embedding vector and the caller intent embedding vector;

[0099] The decision generation module 40 is used to generate decision instructions based on the matching score.

[0100] Figure 3 This is a schematic diagram of the physical structure of an electronic device provided in an embodiment of the present invention, such as... Figure 3As shown, the electronic device 50 includes: a processor 501, a memory 502, and a bus 503;

[0101] The processor 501 and the memory 502 communicate with each other via the bus 503.

[0102] The processor 501 is used to call program instructions in the memory 502 to execute the methods provided in the above-described method embodiments, and to execute the methods provided in the embodiments of the present invention.

[0103] This embodiment provides a non-transitory computer-readable storage medium that stores computer instructions, which cause a computer to execute the method provided in this embodiment of the invention.

[0104] Those skilled in the art will understand that all or part of the steps of the above method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When the program is executed, it performs the steps of the above method embodiments. The aforementioned storage medium includes various storage media capable of storing program code, such as ROM, RAM, magnetic disk, or optical disk.

[0105] 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 occur depending on 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 personalized call handling method, characterized in that, include: Analyze the call intent of unknown calls received by the terminal; Extract the user preference profile embedding vector and caller intent embedding vector of the current user from a pre-trained dynamic heterogeneous graph; Before the steps of extracting the current user's preference profile embedding vector and caller intent embedding vector from the pre-trained dynamic heterogeneous graph, the following steps are included: Construct a node set for a dynamic heterogeneous graph, the node set including user nodes, intent nodes, and entity nodes: Create association edges for each node in the dynamic heterogeneous graph, including user-intent edges, intent-entity edges, and intent-intent edges; Calculate the matching score between the user preference profile embedding vector and the caller intent embedding vector; Decision instructions are generated based on the matching scores.

2. The personalized call processing method according to claim 1, characterized in that, The call intent of the unknown caller received by the parsing terminal includes: After receiving an unknown call from a terminal, the cloud server interacts with the unknown caller to obtain interaction data. The cloud server parses the interaction data to obtain the call intent of the unknown caller.

3. The personalized call processing method according to claim 2, characterized in that, Before the steps of extracting the current user's preference profile embedding vector and caller intent embedding vector from the pre-trained dynamic heterogeneous graph, the following steps are also included: The dynamic heterogeneous graph is input into the heterogeneous graph neural network model to obtain the embedding vector of each node in the dynamic heterogeneous graph; A user preference profile is constructed based on the embedding vector of each node.

4. The personalized call processing method according to claim 1, characterized in that, The personalized call handling method further includes: After the call ends, collect the user's feedback data from this interaction; The heterogeneous graph neural network model is updated and trained based on the feedback data.

5. The personalized call processing method according to claim 1, characterized in that, The calculation of the matching score between the user preference profile embedding vector and the caller intent embedding vector includes: The matching score is calculated using formula (1); Formula (1); In the formula, To match the score, Embed numerical values ​​for each dimension of the vector to create a user preference profile. The numerical values ​​of each dimension of the vector are embedded to represent the caller's intent.

6. The personalized call processing method according to claim 1, characterized in that, The decision instruction generated based on the matching score includes: If the matching score is less than the first preset threshold, a decision instruction to directly intercept is generated; If the matching score is greater than the first preset threshold and less than the second preset threshold, a decision instruction to substitute the connection is generated. If the matching score is greater than the second preset threshold, a decision instruction for transferring the user is generated.

7. A system for personalized call processing, characterized in that, include: The intent parsing module is used to parse the call intent of unknown calls received by the terminal; The vector extraction module is used to extract the user preference profile embedding vector and the caller intent embedding vector of the current user from a pre-trained dynamic heterogeneous graph. Before extracting the current user's preference profile embedding vector and caller intent embedding vector from the pre-trained dynamic heterogeneous graph, the process includes: Construct a node set for a dynamic heterogeneous graph, the node set including user nodes, intent nodes, and entity nodes: Create association edges for each node in the dynamic heterogeneous graph, including user-intent edges, intent-entity edges, and intent-intent edges; The matching calculation module is used to calculate the matching score between the user preference profile embedding vector and the caller intent embedding vector; The decision generation module is used to generate decision instructions based on the matching score.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the method as described in any one of claims 1 to 6.

9. A non-transitory computer-readable medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method as described in any one of claims 1 to 6.

Images