Information extraction method and device, computer device and storage medium

Abstract

The application relates to an information extraction method and device, computer equipment and a storage medium. The method comprises the following steps: obtaining conversation information between a user and a service object; extracting user portrait information and an information type of the user portrait information from the conversation information; determining a position of the user portrait information in the conversation information based on the conversation information and the user portrait information, and obtaining position information of the position; and determining an attribution object of the user portrait information based on the conversation information, the position information and the information type, wherein the attribution object is a person or object to which the user portrait information belongs. The method can improve the accuracy of information extraction.

Description

Technical Field

[0001] This application relates to the field of computer technology, and in particular to an information extraction method, apparatus, computer device, and storage medium. Background Technology

[0002] With the development of computer technology, businesses can use technologies such as artificial intelligence and big data to collect user information and build user profiles. This allows them to obtain more comprehensive and accurate information about users' needs and better serve them. In the process of collecting user information, dependency parsing is typically performed on users' chat messages to extract user information.

[0003] However, users' chat sentences often omit the subject, making it difficult for traditional information extraction methods to accurately extract the information. Summary of the Invention

[0004] Therefore, it is necessary to provide an information extraction method, apparatus, computer equipment, and storage medium that can improve the accuracy of the above-mentioned technical problems.

[0005] An information extraction method, the method comprising:

[0006] Obtain dialogue information between users and service recipients;

[0007] User profile information and the information type of the user profile information are extracted from the dialogue information;

[0008] Based on the dialogue information and the user profile information, determine the position of the user profile information in the dialogue information, and obtain the position information of the position;

[0009] Based on the dialogue information, the location information, and the information type, the belonging object of the user profile information is determined, and the belonging object is the person or thing to which the user profile information belongs.

[0010] An information extraction device, the device comprising:

[0011] The dialogue information acquisition module is used to acquire dialogue information between users and service objects;

[0012] The key information extraction module is used to extract user profile information and the information type of the user profile information from the dialogue information; based on the dialogue information and the user profile information, determine the position of the user profile information in the dialogue information, and obtain the position information of the position;

[0013] The relationship attribution module is used to determine the attribution object of the user profile information based on the dialogue information, the location information, and the information type. The attribution object is the person or thing to which the user profile information belongs.

[0014] A computer device includes a memory and a processor, the memory storing a computer program, the processor executing the computer program to implement the steps of the method described above.

[0015] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method described above.

[0016] The aforementioned information extraction method, apparatus, computer equipment, and storage medium acquire dialogue information between users and service recipients; extract user profile information and information types from the dialogue information; based on the dialogue information and user profile information, the position of user profile information within the dialogue information can be accurately determined, and the position information can be obtained. Furthermore, based on the dialogue information, position information, and information type, the recipient of the user profile information can be accurately determined, avoiding the problem of inaccurate information extraction when the user omits the subject in the dialogue information, thus improving the accuracy of information extraction. Attached Figure Description

[0017] Figure 1 This is a diagram illustrating the application environment of the information extraction method in one embodiment;

[0018] Figure 2 This is a flowchart illustrating an information extraction method in one embodiment;

[0019] Figure 3 This is a flowchart illustrating the steps of filtering related dialogue information from historical dialogue information based on current dialogue information in one embodiment.

[0020] Figure 4 Here is a structural diagram of a BiMPM network in one embodiment;

[0021] Figure 5 Here is a structural diagram of a BiLSTM+CRF model in one embodiment;

[0022] Figure 6 This is a flowchart illustrating the steps of determining the attribution of user profile information based on dialogue information, location information, and information type in one embodiment.

[0023] Figure 7 This is a flowchart illustrating the information extraction process in another embodiment;

[0024] Figure 8 This is a schematic diagram of the interface of an information extraction method in one embodiment;

[0025] Figure 9 This is a structural block diagram of an information extraction device in one embodiment;

[0026] Figure 10 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation

[0027] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0028] The solutions provided in this application relate to technologies such as artificial intelligence and machine learning (ML). Artificial intelligence utilizes digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceiving the environment, acquiring knowledge, and using that knowledge to obtain optimal results, thus enabling machines to possess perception, reasoning, and decision-making capabilities. Machine learning involves multiple disciplines such as probability theory, statistics, approximation theory, convex analysis, and algorithm complexity theory. It studies how computers can simulate or implement human learning behavior to acquire new knowledge or skills and reorganize existing knowledge structures to continuously improve their performance. Machine learning is the core of artificial intelligence and the fundamental way to endow computers with intelligence; its applications span all areas of artificial intelligence. Machine learning and deep learning typically include techniques such as artificial neural networks, belief networks, reinforcement learning, transfer learning, inductive learning, and instructional learning.

[0029] By using technologies such as artificial intelligence and machine learning, more accurate user profile information, information types of user profile information, and the objects to which user profile information belongs can be extracted from dialogue information, thereby achieving more accurate extraction of information from object information.

[0030] The information extraction method provided in this application can be applied to, for example... Figure 1In the application environment shown, the user's terminal 102 communicates with the computer device 104 where the service recipient is located via a network. The user sends a dialogue to the computer device 104 through the terminal 102. The computer device 104 obtains the dialogue information between the user and the service recipient; extracts user profile information and the information type of the user profile information from the dialogue information; determines the position of the user profile information in the dialogue information based on the dialogue information and the user profile information, and obtains the position information; and determines the recipient of the user profile information based on the dialogue information, the position information, and the information type. The terminal 102 can be, but is not limited to, various personal computers, laptops, smartphones, tablets, and portable wearable devices. The computer device 104 can be a terminal or a server. When the computer device 104 is a terminal, it can be, but is not limited to, various personal computers, laptops, smartphones, tablets, and portable wearable devices. When the computer device 104 is a server, it can be implemented using a standalone server or a server cluster composed of multiple servers.

[0031] In one embodiment, such as Figure 2 As shown, an information extraction method is provided, which can be applied to... Figure 1 Taking a computer device as an example, the explanation includes the following steps:

[0032] Step 202: Obtain the dialogue information between the user and the service recipient.

[0033] The service recipient is the entity that provides services to the user. Optionally, the service recipient can be a service person or an intelligent service robot, and is not limited to these. The services provided by the service recipient can be consultation services, business processing services, after-sales services, etc. The service recipient can provide services for a specific industry; for example, in the insurance industry, it can provide insurance application services, claims services, policy cancellation services, consultation services, etc.; in the telecommunications industry, it can provide telecommunications service application services, telecommunications service cancellation services, consultation services, etc.

[0034] During the process of a service provider offering services to a user, the user and the service provider engage in dialogue, and the computer device can acquire the dialogue information between the user and the service provider. This dialogue information can specifically be at least one of the following: text information, image information, video information, audio information, etc.

[0035] In one embodiment, after a computer device obtains dialogue information between a user and a service recipient, it can preprocess the dialogue information and then extract user profile information and the information type of the user profile information from the preprocessed dialogue information.

[0036] Preprocessing can include handling special characters, converting English characters to uppercase and lowercase, and unifying simplified and traditional Chinese characters. Handling special characters can involve deleting them, converting them to ordinary characters, marking them, and so on.

[0037] Preprocessing may also include using regular expressions to match email addresses, URLs, phone numbers, ID numbers, etc., and replacing them with specially marked text.

[0038] Step 204: Extract user profile information and the information type of user profile information from the dialogue information.

[0039] User profiling refers to abstracting specific information about a user into tags, using these tags to concretize the user's image, thereby providing targeted services. User profiling information can include user behavior information and user attribute information. User behavior information specifically includes browsing information, purchase information, and preference information. User attribute information specifically includes age, gender, height, weight, skin color, and place of residence.

[0040] The information type of user profile information refers to the type to which the user profile information belongs. For example, the information type of user profile information "male" is "gender", the information type of user profile information "25" is "age", the information type of user profile information "180cm" is "height", the information type of user profile information "watching article A for 5 minutes" is "browsing information", and the information type of user profile information "purchasing product B" is "purchasing information".

[0041] Step 206: Based on the dialogue information and user profile information, determine the position of the user profile information in the dialogue information and obtain the position information.

[0042] User profile information is extracted from dialogue information. Therefore, computer devices can determine the position of the user profile in the dialogue information based on the position extracted from the dialogue information, and obtain the position information of that position.

[0043] In one implementation, the computer device calculates the string length between the current position and the beginning of the dialogue message. The string length represents the position of the user profile information within the dialogue message. For example, in the dialogue message "How old are your family members? [User] Wife 43, two children 3 and 8 years old," the user profile information is 43. The computer device determines the position of the user profile information in the dialogue message as offset: 20, meaning the string length between the position of user profile information "43" and the beginning of the dialogue message is 20. As another example, in the dialogue message "How old are you and your husband, and your children? [User] Husband 38, me 34, eldest 5 and a half, youngest less than one year old," the user profile information is 34. The computer device determines the position of the user profile information in the dialogue message as offset: 27, meaning the string length between the position of user profile information "34" and the beginning of the dialogue message is 27.

[0044] In another implementation, the computer device calculates the string length between the current position and the end of the dialogue message. The string length represents the position of the user profile information within the dialogue message. For example, in the dialogue message "How old are your family members? [User] Wife 43, two children 3 and 8 years old," the user profile information is 43. The computer device determines the position of the user profile information within the dialogue message as offset: 10, indicating that the string length between the position of the user profile information "43" and the end of the dialogue message is 10.

[0045] Step 208: Based on dialogue information, location information, and information type, determine the attribution object of the user profile information. The attribution object is the person or thing to which the user profile information belongs.

[0046] The attribution object refers to the person or thing to which the user profile information belongs. For example, in the dialogue message "How old are your family members? [User] Wife 43, two children 3 and 8 years old", the attribution object for the user profile information 43 is the wife, and the attribution object for the user profile information 3 and 8 years old is the two children. As another example, in the dialogue message "How much does the refrigerator weigh? [User] 10 jin", the attribution object for the user profile information 10 jin is the refrigerator.

[0047] It is understandable that users often omit the subject in their dialogue with the service recipient. For example, "How old is your wife? [User] 43". The user profile information extracted from the dialogue is 43, the information type is age, and the location information of the user profile information is determined to be offset: 16. Based on the dialogue information, location information, and information type, it can be determined that the user profile information belongs to the wife.

[0048] Specifically, the computer device inputs dialogue information, location information, and information type into the trained relationship attribution model, and determines the attribution object of the user profile information through the trained relationship attribution model.

[0049] In this embodiment, dialogue information between the user and the service recipient is acquired; user profile information and information type of the user profile information are extracted from the dialogue information. Based on the dialogue information and user profile information, the position of the user profile information in the dialogue information can be accurately determined, and the position information is obtained. Then, based on the dialogue information, position information, and information type, the recipient of the user profile information can be accurately determined, avoiding the problem of inaccurate information extraction when the user omits the subject in the dialogue information, thus improving the accuracy of information extraction. Furthermore, the above embodiment, considering the interactive and spoken characteristics of user chat, can more effectively mine the information included in the dialogue information from multiple levels.

[0050] In one embodiment, the dialogue information includes current dialogue information and associated dialogue information; obtaining dialogue information between users and service objects includes: obtaining current dialogue information between users and service objects, as well as historical dialogue information corresponding to the current dialogue information; and filtering out associated dialogue information related to the current dialogue information from the historical dialogue information based on the current dialogue information.

[0051] Current dialogue information refers to the dialogue between the user and the service recipient at that time. Historical dialogue information refers to dialogue information that occurred before the current dialogue information. Related dialogue information refers to dialogue information in the historical dialogue information that is related to the current dialogue information.

[0052] In one implementation, the computer device acquires current dialogue information within the current preset window and acquires historical dialogue information from each preset window preceding the current preset window. For example, the dialogue information in the current preset window is User_n, and the historical dialogue information from each preset window preceding the current preset window is User_n-1, User_n-2, ..., User_1.

[0053] In another implementation, the computer device acquires the latest specified number of dialogue messages as the current dialogue messages, and acquires the information preceding the current dialogue messages in chronological order as historical dialogue messages. The specified number can be set as needed.

[0054] The computer device concatenates the current dialogue information and related dialogue information in chronological order to obtain dialogue information, and then extracts user profile information and the information type of the user profile information from the dialogue information. Optionally, the chronological order can be from early to late or from late to early.

[0055] It is understandable that current and historical dialogue information may be from the same round of dialogue between the user and the service recipient, so there must be a correlation between them. However, historical dialogue information may also contain some noise information that is unrelated to the current dialogue information. Therefore, based on the current dialogue information, we can filter out related dialogue information from historical dialogue information and filter out noise information that is unrelated to the current dialogue information. This allows us to extract user information more accurately and quickly based on the current and related dialogue information.

[0056] In one embodiment, such as Figure 3 As shown, based on the current dialogue information, related dialogue information relevant to the current dialogue information is filtered from the historical dialogue information, including:

[0057] Step 302: Perform vector transformation on the current dialogue information to obtain the current dialogue vector; perform vector transformation on the historical dialogue information to obtain the historical dialogue vector.

[0058] The current dialogue vector is a vector representing information from the current dialogue. The historical dialogue vector is a vector representing information from the previous dialogue.

[0059] Specifically, the computer device uses a trained word vector model to segment the current dialogue information or historical dialogue vectors into words, transforms each word into a vector, and then concatenates these word vectors to obtain the current dialogue vector or historical dialogue vector. Both the current dialogue vector and the historical dialogue vector are sentence vectors. The word vector model can specifically be a Skip-Gram model.

[0060] The training method for a word vector model includes: acquiring various proper nouns, integrating these proper nouns into a word segmentation dictionary, segmenting the training corpus using the word segmentation dictionary, and inputting the segmented training corpus into the word vector model for training, resulting in a trained word vector model. The computer device can use a word segmentation dictionary and the forward maximum matching method to segment the training corpus.

[0061] Proper terms can be specific terms used in various industries, such as proper nouns in the insurance industry (e.g., insurance application, claims, illness, type of insurance, insurance name), and proper nouns in the telecommunications industry (e.g., server, switch).

[0062] Training corpora can include dialogues between other users and service recipients, as well as text from various documents or knowledge bases. For example, training corpora could include insurance-related documents from the insurance industry, text from insurance knowledge bases, and dialogues between other users and insurance service recipients.

[0063] If computer equipment uses various proper nouns to train the word vector model, the trained word vector model can more quickly convert dialogue information containing various proper nouns into vectors, reducing the cost of manual annotation.

[0064] Step 304: Concatenate the current dialogue vector and the historical dialogue vector to obtain the concatenated dialogue vector.

[0065] Optionally, the computer device can concatenate the current dialogue vector to the historical dialogue vector to obtain a concatenated dialogue vector, or it can concatenate the historical dialogue vector to the current dialogue vector to obtain a concatenated dialogue vector. For example, if the current dialogue vector is A and the historical dialogue vector is B, the computer device can concatenate the current dialogue vector A and the historical dialogue vector B to obtain a concatenated dialogue vector (A,B) or (B,A).

[0066] Step 306: Determine the semantic similarity between the current dialogue information and each sub-dialogue information in the historical dialogue information based on the concatenated dialogue vector. Based on the semantic similarity, filter out the related dialogue information that is relevant to the current dialogue information from the historical dialogue information.

[0067] The computer device inputs the concatenated dialogue vectors into a trained context-related text pair matching network. This network determines the semantic similarity between the current dialogue information and each sub-dialogue information in the historical dialogue information, and then filters out related dialogue information from the historical dialogue information. The context-related text pair matching network is specifically designed to filter out relevant text within the context. This network can be a Bilateral Multi-Perspective Matching (BiMPM) network.

[0068] Specifically, the keyword feature information of the current dialogue information and the keyword feature information of each sub-dialogue information in the historical dialogue information are extracted by the context-related text matching network. The semantic similarity between the current dialogue information and each sub-dialogue information in the historical dialogue information is determined by using the keyword feature information as the main feature.

[0069] Computer devices can consider sub-dialogue information with a semantic similarity higher than a preset similarity as related dialogue information. The preset similarity can be set as needed. For example, if the preset similarity is 50%, the semantic similarity between the current dialogue information and sub-dialogue information A is 40%, the semantic similarity between the current dialogue information and sub-dialogue information B is 56%, and the semantic similarity between the current dialogue information and sub-dialogue information C is 80%, then sub-dialogue information B and sub-dialogue information C are both related dialogue information associated with the current dialogue information.

[0070] Figure 4 This is a structural diagram of a BiMPM network in one embodiment. The BiMPM network includes a word representation layer, a context representation layer, a matching layer, an aggregation layer, and a prediction layer. The computer device inputs the concatenated dialogue vector into the BiMPM network, which processes it sequentially through the word representation layer, context representation layer, matching layer, aggregation layer, and prediction layer, filtering out relevant dialogue information related to the current dialogue information from historical dialogue information.

[0071] For the concatenated dialogue vector obtained by concatenating the current dialogue vector and the historical dialogue vector, word segmentation is performed on each sub-dialogue information in the corresponding current dialogue information and historical dialogue information. Specifically, in the word representation layer, the current dialogue information is represented as words p1, p2, p3, ..., pM. Any sub-dialogue information is represented as words q1, q2, q3, ..., qN. The word representation layer inputs the word segmented current dialogue information and sub-dialogue information into the context representation layer. The context representation layer extracts the word order relationship between any two adjacent words in the current dialogue information and sub-dialogue information and represents it using a word order representation vector. Specifically, the word order relationship between any two adjacent words can be a word order relationship from the beginning to the end of the sentence or a word order relationship from the end to the beginning of the sentence. The matching layer matches the word order representation vector of the current dialogue information with the word order representation vector of any two adjacent words in the sub-dialogue information and outputs a matching vector. The matching layer inputs the matching vectors into the aggregation layer, which aggregates the individual matching vectors according to word order to obtain an aggregated vector. This aggregated vector is then input into the prediction layer, which predicts the semantic similarity between the current dialogue information and the sub-dialogue information based on these aggregated vectors.

[0072] In this embodiment, the current dialogue information and the historical dialogue information are transformed into vectors respectively. Then, the current dialogue vector and the historical dialogue vector are concatenated to obtain the concatenated dialogue vector. Based on the concatenated dialogue vector, the semantic similarity between the current dialogue information and each sub-dialogue information in the historical dialogue information is determined. Based on the semantic similarity between the current dialogue information and each sub-dialogue information in the historical dialogue information, the related dialogue information can be accurately filtered out from the historical dialogue information, while noise information unrelated to the current dialogue information is filtered out. Thus, user profile information can be extracted more accurately and faster from the current dialogue information and related dialogue information.

[0073] In one embodiment, the training method of the context-related text pair matching network includes: obtaining dialogue training text; the dialogue training text includes the current dialogue training text and historical dialogue training text, the historical dialogue training text includes positive training text related to the current dialogue training text and negative training text unrelated to the current dialogue training text; inputting the current dialogue training text, positive training text and negative training text into the context-related text pair matching network, training the context-related text pair matching network, and obtaining the trained context-related text pair matching network.

[0074] Dialogue training text is the dialogue text used for training. Dialogue training text can be historical dialogue information between the user being extracted and the service recipient, or historical dialogue information between other users and the service recipient, and is not limited to these.

[0075] A computer device can set a specific text from the dialogue training text as the current dialogue training text, and then set the text preceding the current dialogue training text as the historical dialogue training text. From the historical dialogue training text, texts related to the current dialogue training text are set as positive training texts, and texts unrelated to the current dialogue training text are set as negative training texts. The current dialogue training text, positive training texts, and negative training texts are then input into a context-related text pair matching network. Machine learning can be used to train the context-related text pair matching network, resulting in a trained context-related text pair matching network.

[0076] In this embodiment, the current dialogue training text, positive training text related to the current dialogue training text, and negative training text unrelated to the current dialogue training text are input into the context-related text pair matching network. This can train a more accurate context-related text pair matching network, thereby enabling the selection of more accurate related dialogue information in the subsequent information extraction process.

[0077] In one embodiment, extracting user profile information and the information type of user profile information from dialogue information includes: performing vector transformation on the dialogue information to obtain dialogue vectors; labeling the dialogue vectors to obtain labeling information; and decoding each labeling information to extract user profile information and the information type of user profile information.

[0078] Labeling information can include one or more of the following: B-LOC, I-LOC, B-Person, I-Person, B-Organization, I-Organization, and O. B-LOC represents the beginning of a location, I-LOC represents the middle part of a location, B-Person represents the beginning of a person's name, I-Person represents the middle part of a person's name, B-Organization represents the beginning of an organization's name, I-Organization represents the middle part of an organization's name, and O represents non-entity information.

[0079] In one implementation, the computer device can use a trained word vector model to perform vector transformation on the dialogue information to obtain dialogue vectors. These dialogue vectors are then input into a BiLSTM+CRF model, which annotates the vectors to obtain annotation information. The BiLSTM+CRF model includes BiLSTM (Bi-directional Long-Short Term Memory) layers and CRF layers. The BiLSTM layers are particularly effective at capturing longer-distance dependencies within the dialogue vectors.

[0080] Specifically, the computer device inputs the dialogue vectors into the BiLSTM+CRF model, and annotates the dialogue vectors through the BiLSTM+CRF model to obtain annotation information. This includes: the computer device inputs the dialogue vectors into the BiLSTM layer, obtains the dependencies in the dialogue vectors through the BiLSTM layer, and then inputs the dialogue vectors carrying the dependencies into the CRF layer, and annotates the dialogue vectors through the CRF layer to obtain annotation information.

[0081] Figure 5 This is a structural diagram of a BiLSTM+CRF model in one embodiment. The computer device inputs the dialogue vector corresponding to "China is very big" into the BiLSTM+CRF model, and the BiLSTM+CRF model annotates the dialogue vector. The dialogue vector can be represented using one-hot vectors.

[0082] The computer equipment then decodes each labeled information to extract user profile information and the information type of the user profile information from the dialogue information. For example, if the labeled information is "O",O","B-family member","I-family member","O","O","B-family member","I-family member","B-age","I-age","O",O", decoding the labeled information, the beginning part of the family member can be extracted from "B-family member", and the middle part of the family member can be extracted from "I-family member", and the information type is family member. That is, "B-family member" and "I-family member" together form the word content of the family member, and the information type is family member. Similarly, the beginning part of the age can be extracted from "B-age", and the information type is age. The middle part of the age can be extracted from "I-age", and the information type is age. That is, "B-age" and "I-age" together form the word content of the age, and the information type is age.

[0083] The computer device can input the dialogue vector into the BiLSTM layer. In other implementations, the dialogue vector can also be input into a Convolutional Neural Network (CNN) or a transformer model.

[0084] Computer devices can input dialogue vectors carrying dependencies into a CRF layer for annotation. In other implementations, dialogue vectors carrying dependencies can also be input into a BERT (Bidirectional Encoder Representations from Transformers) model for annotation.

[0085] In this embodiment, the dialogue information is vectorized to obtain a dialogue vector; the dialogue vector is labeled to obtain labeled information; and the labeled information is decoded to accurately extract user profile information and the information type of user profile information.

[0086] In one embodiment, determining the attribution of user profile information based on dialogue information, location information, and information type includes: performing vector embedding processing on the dialogue information, location information, and information type respectively to obtain dialogue vector, location vector, and type vector; concatenating the dialogue vector, location vector, and type vector to obtain an information extraction vector; performing dense processing on the information extraction vector; performing logistic regression processing on the densed information extraction vector to obtain the probability that the user profile information belongs to each candidate object; and predicting the attribution of the user profile information based on the probability that the user profile information belongs to each candidate object.

[0087] Vector embedding refers to the process of converting text into vectors. A dialogue vector is a vector obtained by embedding dialogue information. A type vector is a vector obtained by embedding information types. The information extraction vector is a vector obtained by concatenating the dialogue vector, position vector, and type vector. Logistic regression is a generalized form of linear regression analysis, commonly used in data mining, automated disease diagnosis, and economic forecasting.

[0088] A location vector is a vector obtained by embedding location information. It represents the relative position of the entity containing the user profile information to other entities in the dialogue information. An entity is text with actual meaning. For example, an entity could be "age," "Shenzhen," "we," "34," etc. Typically, dialogue information includes multiple entities, and each dimension of the location vector represents the relative position of the entity containing the user profile information to other entities. The relative position can be the number of strings between the entity containing the user profile information and other entities.

[0089] Computer devices use word vector models to compress high-dimensional information extraction vectors into low-dimensional dense vectors. These dense vectors are then subjected to logistic regression to obtain the probability that the user profile information belongs to each candidate object. This probability is then input into a feedforward network, which determines the user profile information's belonging object from among the candidate objects. Specifically, the word vector model can be a Word2Vec model. The value of the dense vector is simply a double array. Candidate objects can be any person or thing in the dialogue information. Furthermore, candidate objects can also be any person or thing in the sentence containing the user's information.

[0090] Dense vectors are defined in contrast to sparse vectors obtained after one-hot encoding. Sparse vectors have a high dimensionality, but most elements are zero; dense vectors, on the other hand, have most elements that are non-zero. Word vector models project the words encoded by sparse vectors into a low-dimensional space to obtain denser vectors, which are more dense than sparse vectors.

[0091] In one implementation, a feedforward network can be used to determine the candidate with the highest probability as the recipient of the user profile information. In another implementation, a feedforward network can be used to determine the candidate with the second highest probability as the recipient of the user profile information. The specific method of determining the recipient of the user profile information using a feedforward network is not limited.

[0092] In this embodiment, the dialogue information, location information, and information type are respectively processed by vector embedding to obtain dialogue vector, location vector, and type vector; the dialogue vector, location vector, and type vector are concatenated to obtain the information extraction vector; the information extraction vector is densified, and the densified information extraction vector is subjected to logistic regression processing to obtain the probability that the user profile information belongs to each candidate object. Based on the probability that the user profile information belongs to each candidate object, the vector can be extracted more accurately to predict the belonging object of the user profile information.

[0093] In one embodiment, concatenating the dialogue vector, location vector, and type vector to obtain the information extraction vector includes: concatenating the dialogue vector and location vector to obtain a dialogue vector carrying location information; encoding the dialogue vector carrying location information to obtain an encoded vector; and concatenating the encoded vector and type vector to obtain the information extraction vector.

[0094] An encoding vector is a vector obtained by encoding a dialogue vector that carries location information.

[0095] Optionally, the encoding vector can be concatenated before or after the type vector, and is not limited thereto. For example, if the current dialogue vector is A and the historical dialogue vector is B, the computer device can concatenate the current dialogue vector A and the historical dialogue vector B to obtain either the concatenated dialogue vector (A,B) or (B,A).

[0096] When a computer device concatenates a dialogue vector and a location vector, the resulting dialogue vector contains the location information corresponding to the location vector. Therefore, the concatenated dialogue vector carries location information.

[0097] The computer device uses a trained encoding model to encode the dialogue vector carrying location information, resulting in an encoded vector. The encoding model can be one of the following: a bidirectional RNN (Recurrent Neural Network), an LSTM (Long Short-Term Memory) network, or a GRU (Gated Recurrent Unit).

[0098] In this embodiment, the dialogue vector and the location vector are concatenated to obtain a dialogue vector carrying location information; the dialogue vector carrying location information is encoded to obtain an encoded vector; the encoded vector and the type vector are concatenated to obtain an information extraction vector, which includes location information, thereby enabling more accurate extraction of the user profile information's belonging object.

[0099] Figure 6This is a flowchart illustrating the steps for determining the attribution of user profile information based on dialogue information, location information, and information type in one embodiment. The computer device performs vector embedding processing on the dialogue information, location information, and information type to obtain dialogue vectors, location vectors, and type vectors, respectively. The dialogue vectors and location vectors are concatenated to obtain a dialogue vector carrying location information. This dialogue vector carrying location information is input into an encoding model, which encodes the sentence vectors to obtain an encoded vector. The encoded vector and type vector are concatenated to obtain an information extraction vector. The information extraction vector is then densified to obtain a dense vector. Logistic regression is performed on the dense vector, and the attribution of the user profile information is predicted based on the information extraction vector after logistic regression processing.

[0100] In one embodiment, the method further includes displaying user profile information, information type, and belonging object in the user information display area of ​​the terminal interface of the service object.

[0101] When the computer device is a terminal, the terminal interface of the service recipient includes a user information display area, which can display user profile information, information type, and the recipient.

[0102] When the computer device is a server, the server sends user profile information, information type, and belonging object to the terminal where the service object is located. The terminal interface includes a user information display area, which can display user profile information, information type, and belonging object.

[0103] The user information display area can be one of the following: the right side, left side, or top side of the terminal interface.

[0104] Computer devices display user profile information, information type, and belonging object in the terminal interface of the service recipient, so that the service recipient can quickly obtain the user profile information, information type, and belonging object when serving the user, thereby providing better service to the user.

[0105] Among these, user profile information, information type, and target audience are all structured information. Structured information refers to information that has been analyzed and broken down into multiple interconnected components, with a clear hierarchical structure among these components.

[0106] In one embodiment, the method further includes: performing similarity matching between the processing object and each preset text, and determining the target object that matches the processing object from each preset text, wherein the processing object includes any one of user profile information, information type and belonging object, and the target object is any one of target user profile information, target information type and target belonging object.

[0107] The preset text can be set as needed.

[0108] In one implementation, the computer device uses a literal similarity function to perform similarity matching between the processing object and various preset texts, determining the target object that matches the processing object from among the preset texts. Specifically, when the Jaccard distance or edit distance between the processing object and the preset text is less than a preset distance, the processing object matches the preset text, and the preset text is taken as the target object. The Jaccard distance is a metric used to measure the difference between two sets. The edit distance is a quantitative measure of the difference between two strings (e.g., English words), measured by the minimum number of processing steps required to transform one string into another.

[0109] In another implementation, the computer device uses a shallow semantic similarity function to perform similarity matching between the object to be processed and each preset text, and determines the target object that matches the object to be processed from each preset text.

[0110] For example, different user profile information may be "healthy", "good", "very good", or "very healthy". Computer devices can normalize the different user profile information to obtain a unified user profile information of "healthy".

[0111] In this embodiment, a literal similarity function is used to match the processing object with each preset text. The target object that matches the processing object is determined from each preset text. User profile information, information type and belonging object can be normalized respectively. Sentences or words with the same intent but different expressions are unified into the same target object, which can be used more quickly for subsequent data filtering and use.

[0112] In one embodiment, the target user profile information, target information type, and target belonging object are displayed in the user information display area of ​​the terminal interface of the service recipient.

[0113] When the computer device is a terminal, the terminal interface of the service recipient includes a user information display area, which can display target user profile information, target information type, and target belonging object.

[0114] When the computer device is a server, the server sends the target user profile information, target information type, and target belonging object to the terminal where the service object is located. The terminal interface includes a user information display area, which can display the target user profile information, target information type, and target belonging object.

[0115] The user information display area can be one of the following: the right side, left side, or top side of the terminal interface.

[0116] In another embodiment, when the computer device extracts numerical information from user profile information, information type, and belonging object, it uses regular expressions to transform the numerical information to obtain target information, which is then displayed on the terminal interface of the service recipient. For example, if the user profile information indicates birth in 1994, using regular expressions to transform the user profile information will yield the user's age as 26 years old.

[0117] Figure 7 This is a flowchart illustrating information extraction in another embodiment. A computer device provides an information extraction apparatus, including a dialogue information acquisition module, a key information extraction module, a relationship attribution module, and a normalization module. The computer device acquires current and historical dialogue information between the user and the service recipient. It performs text preprocessing on both the current and historical dialogue information, and then inputs the preprocessed current and historical dialogue information into a trained word vector model to obtain current and historical dialogue vectors, respectively. The computer device can pre-train the word vector model using insurance documents and documents from a knowledge base to obtain a trained word vector model.

[0118] The computer device concatenates the current dialogue vector and the historical dialogue vector to obtain the concatenated dialogue vector. Then, through a trained context-related text pair matching network, it filters out related dialogue information from the historical dialogue information that is relevant to the current dialogue information.

[0119] The computer device inputs the current dialogue information and related dialogue information into the encoder, and then inputs the encoder's output into the CRF layer to extract user profile information and its information type. The encoder can be a bidirectional long short-term memory network (LSTM), which extracts the dependencies between the current and related dialogue information. The current dialogue information carrying these dependencies, along with the related dialogue information carrying these dependencies, is then input into the CRF layer to extract user profile information and its information type. Alternatively, the encoder can be a convolutional neural network (CNN) or a transformer model.

[0120] Based on dialogue information and user profile information, the computer device determines the position of the user profile information within the dialogue information, obtains the position information, and then inputs the current dialogue information, related dialogue information, position information, and information type into a relation extraction network, outputting the object to which the user profile information belongs. Specifically, the relation extraction network sequentially embeds, encodes, loads, and predicts the current dialogue information, related dialogue information, position information, and information type to determine the object to which the user profile information belongs.

[0121] The computer device performs similarity matching between the object to be processed and each preset text, and determines the target object that matches the object to be processed from each preset text. The object to be processed includes any one of user profile information, information type and belonging object, and the target object is any one of target user profile information, target information type and target belonging object. The target user profile information, target information type and target belonging object are displayed in the terminal interface of the service object.

[0122] In one embodiment, another information extraction method is provided, applied to the terminal of the service object, including the following steps: initiating an information extraction request to the server; obtaining user profile information, information type of user profile information, and belonging object of user profile information sent by the server; wherein, the belonging object is the person or thing to which the user profile information belongs, the belonging object is determined by the server based on the dialogue information, location information, and information type between the user and the service object, the location information is obtained based on the position of the user profile information in the dialogue information, the location is determined based on the dialogue information and the user profile information, the user profile information and information type are extracted from the dialogue information, the dialogue information is obtained based on the information extraction request; and displaying the user profile information, information type, and belonging object in the user information display area.

[0123] Information retrieval requests may include dialogue information, or attribute information of the dialogue information, but are not limited to these. Attribute information includes, for example, the number of dialogue messages and the time range.

[0124] In one implementation, the terminal receives a user's trigger selection of candidate information, determines the candidate information as dialogue information, generates an information retrieval request including the dialogue information, and sends the information retrieval request to the server. The server retrieves the dialogue information from the information retrieval request.

[0125] In another implementation, the terminal sends an information retrieval request, including attribute information, to the server. Based on the attribute information in the information retrieval request, the server retrieves the corresponding dialogue information from its memory.

[0126] Optionally, in the user information display area, the terminal may arrange and display user profile information, information type, and belonging object, or it may display user profile information, information type, and belonging object in a table format, or it may not be limited to this.

[0127] In this embodiment, the terminal initiates an information retrieval request to the server. Based on the information retrieval request, the server determines the dialogue information between the user and the service object, extracts user profile information and the information type of the user profile information from the dialogue information, and accurately determines the position of the user profile information in the dialogue information based on the dialogue information and the position information. Then, based on the dialogue information, position information, and information type, the user profile information's owner can be accurately determined, avoiding the problem of inaccurate information retrieval when the user omits the subject in the dialogue information, thus improving the accuracy of information retrieval. The user profile information, information type, and owner are then sent to the terminal. The terminal can display the user profile information, information type, and owner in the user information display area.

[0128] In one embodiment, such as Figure 8 As shown, the visual interface of the service recipient's terminal includes a dialogue area 802 and a user information display area 804. The visual interface can be the interface of a social product or a temporary work session interface. The dialogue area of ​​the visual interface displays the dialogue information between the user and the service recipient, thereby extracting user profile information, information type, and the object to which the user profile information belongs from the dialogue information between the user and the service recipient, and displaying it in the user information display area. The display format includes conversational, tabular, and tag-based formats. In a specific implementation process, the computer device obtains the dialogue information between the user and the service recipient in the chat window: "How old are you and your husband, and how old are your children?", "My husband is 38, I am 34, our eldest is 5 and a half, and our youngest is less than one year old," "Are you and your husband in good health?", "How is your child's immune system?", "Very good." From the above dialogue information, user profile information, the information type of the user profile information, and the user profile's target audience can be extracted. User profile information includes 61, where the information type of 61 is age, and the target audience for 61 is the elderly; user profile information includes children, where the information type of children is family members, and the target audience for children is the user; user profile information includes 34, where the information type of 34 is age, and the target audience for 34 is the user; user profile information includes 38, where the information type of 38 is age, and the target audience for 38 is the spouse; user profile information includes "5 and a half, one year old", where the information type of "5 and a half, one year old" is age, and the target audience for "5 and a half, one year old" is children; user profile information includes "Very good", where the information type of "Very good" is age, and the target audience for "Very good" is children.

[0129] The computer device will display the extracted user profile information, the information type of the user profile, and the object to which the user profile belongs on the interface. This allows the service provider to quickly and directly understand the user's situation and needs when serving one or more users, providing services to the user more quickly, improving response efficiency, and ensuring that the response content is targeted.

[0130] It should be understood that, although Figure 2 , Figure 3 , Figure 6 and Figure 7 The steps in the flowchart are shown sequentially as indicated by the arrows, but these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order in which these steps are executed, and they can be performed in other orders. Figure 2 , Figure 3 , Figure 6 and Figure 7 At least some of the steps in the process may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but may be executed at different times. The execution order of these steps or stages is not necessarily sequential, but may be executed in turn or alternately with other steps or at least some of the steps or stages in other steps.

[0131] In one embodiment, such as Figure 9 As shown, an information extraction device is provided. This device can be a software module, a hardware module, or a combination of both as part of a computer device. Specifically, the device includes: a dialogue information acquisition module 902, a key information extraction module 904, and a relationship attribution module 906, wherein:

[0132] The dialogue information acquisition module 902 is used to acquire dialogue information between users and service objects.

[0133] The key information extraction module 904 is used to extract user profile information and the information type of user profile information from the dialogue information; based on the dialogue information and user profile information, it determines the position of the user profile information in the dialogue information and obtains the position information.

[0134] The relationship attribution module 906 is used to determine the attribution object of user profile information based on dialogue information, location information, and information type. The attribution object is the person or thing to which the user profile information belongs.

[0135] The aforementioned information extraction device acquires dialogue information between users and service recipients; it extracts user profile information and information type from the dialogue information. Based on the dialogue information and user profile information, it can accurately determine the position of the user profile information in the dialogue information, obtain the position information, and then accurately determine the recipient of the user profile information based on the dialogue information, position information, and information type. This avoids the problem of not being able to accurately extract information when the user omits the subject in the dialogue information, and can improve the accuracy of information extraction.

[0136] In one embodiment, the dialogue information includes current dialogue information and associated dialogue information; the dialogue information acquisition module 902 is further configured to acquire current dialogue information between the user and the service object, as well as historical dialogue information corresponding to the current dialogue information; and based on the current dialogue information, to filter out associated dialogue information related to the current dialogue information from the historical dialogue information.

[0137] In one embodiment, the dialogue information acquisition module 902 is further configured to perform vector conversion on the current dialogue information to obtain the current dialogue vector; perform vector conversion on the historical dialogue information to obtain the historical dialogue vector; concatenate the current dialogue vector and the historical dialogue vector to obtain the concatenated dialogue vector; and based on semantic similarity, based on the semantic similarity between the current dialogue information and each sub-dialogue information in the historical dialogue information, filter out the related dialogue information that is relevant to the current dialogue information from the historical dialogue information.

[0138] In one embodiment, the information extraction device further includes a training module for acquiring dialogue training text; the dialogue training text includes current dialogue training text and historical dialogue training text, the historical dialogue training text includes positive training text related to the current dialogue training text and negative training text unrelated to the current dialogue training text; the current dialogue training text, positive training text and negative training text are input into the context-related text pair matching network, the context-related text pair matching network is trained, and the trained context-related text pair matching network is obtained.

[0139] In one embodiment, the key information extraction module 904 is further configured to perform vector conversion on the dialogue information to obtain a dialogue vector; to annotate the dialogue vector to obtain annotation information; and to decode each annotation information to extract user profile information and the information type of the user profile information.

[0140] In one embodiment, the relationship attribution module 906 is further configured to perform vector embedding processing on the dialogue information, location information, and information type respectively to obtain dialogue vector, location vector, and type vector; concatenate the dialogue vector, location vector, and type vector to obtain information extraction vector; perform dense processing on the information extraction vector; perform logistic regression processing on the densed information extraction vector to obtain the probability that the user profile information belongs to each candidate object; and predict the attribution object of the user profile information based on the probability that the user profile information belongs to each candidate object.

[0141] In one embodiment, the relationship attribution module 906 is further configured to concatenate the dialogue vector and the location vector to obtain a dialogue vector carrying location information; encode the dialogue vector carrying location information to obtain an encoded vector; and concatenate the encoded vector and the type vector to obtain an information extraction vector.

[0142] In one embodiment, the information extraction device further includes a display module for displaying user profile information, information type, and belonging object in the user information display area of ​​the terminal interface of the service recipient.

[0143] In one embodiment, the information extraction device further includes a normalization module, which is used to perform similarity matching between the processing object and each preset text, and to determine the target object that matches the processing object from each preset text. The processing object includes any one of user profile information, information type and belonging object, and the target object is any one of target user profile information, target information type and target belonging object.

[0144] In this embodiment, the information extraction device includes a dialogue information acquisition module, a key information extraction module, a relationship attribution module, and a normalization module. Developers can expand based on a certain module and expand the extracted information for a certain module without having to make overall improvements to the entire device, thus expanding the information extraction device more efficiently.

[0145] In one embodiment, another information extraction device is provided. This device may be a software module, a hardware module, or a combination of both, integrated into a computer device. Specifically, the device includes: a request initiation module, an information acquisition module, and a display module, wherein:

[0146] The request initiation module is used to send information retrieval requests to the server.

[0147] The information acquisition module is used to acquire user profile information, information type of user profile information, and the object to which user profile information belongs, sent by the server. The object to which user profile information belongs is the person or thing to which user profile information belongs. The object to which user profile information belongs is determined by the server based on the dialogue information, location information, and information type between the user and the service object. The location information is obtained based on the position of user profile information in the dialogue information. The location is determined based on the dialogue information and user profile information. User profile information and information type are extracted from the dialogue information. The dialogue information is obtained based on the information acquisition request.

[0148] The display module is used to display user profile information, information type, and belonging object in the user information display area.

[0149] The aforementioned information extraction device involves the terminal initiating an information extraction request to the server. Based on this request, the server determines the dialogue information between the user and the service recipient, extracts user profile information and its information type from the dialogue information, and accurately determines the position of the user profile information within the dialogue information. This positional information, along with the dialogue information, positional information, and information type, accurately determines the recipient of the user profile information. This avoids the problem of inaccurate information extraction when the user omits the subject in the dialogue information, thus improving the accuracy of information extraction. The user profile information, information type, and recipient are then sent to the terminal. The terminal can then display the user profile information, information type, and recipient in the user information display area.

[0150] Specific limitations regarding the information extraction device can be found in the limitations of the information extraction method described above, and will not be repeated here. Each module in the aforementioned information extraction device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device in hardware form, or stored in the memory of a computer device in software form, so that the processor can call and execute the operations corresponding to each module.

[0151] In one embodiment, a computer device is provided, which may be a terminal, and its internal structure diagram may be as follows: Figure 10As shown, the computer device includes a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. When the computer program is executed by the processor, it implements an information retrieval method. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad mounted on the computer device casing, or an external keyboard, touchpad, or mouse.

[0152] Those skilled in the art will understand that Figure 10 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0153] In one embodiment, a computer device is also provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the above method embodiments.

[0154] In one embodiment, a computer-readable storage medium is provided storing a computer program that, when executed by a processor, implements the steps in the above method embodiments.

[0155] In one embodiment, a computer program product or computer program is provided, the computer program product or computer program including computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium, and executes the computer instructions, causing the computer device to perform the steps in the above method embodiments.

[0156] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the methods described above. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, or optical storage, etc. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc.

[0157] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0158] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the invention patent. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this patent application should be determined by the appended claims.

Claims

1. An information extraction method, characterized in that, The method includes: Acquire dialogue information between users and service objects; the dialogue information includes current dialogue information and related dialogue information; acquiring the dialogue information between users and service objects includes: Obtain the current dialogue information between the user and the service object, as well as the historical dialogue information corresponding to the current dialogue information; The current dialogue information is vectorized to obtain the current dialogue vector; the historical dialogue information is vectorized to obtain the historical dialogue vector. The current dialogue vector and the historical dialogue vector are concatenated to obtain the concatenated dialogue vector; The concatenated dialogue vector is input into the trained context-related text pair matching network to obtain the semantic similarity between the current dialogue information and each sub-dialogue information in the historical dialogue information, and to filter out the related dialogue information related to the current dialogue information from the historical dialogue information; the context-related text pair matching network is a BiMPM network. User profile information and the information type of the user profile information are extracted from the dialogue information; Based on the dialogue information and the user profile information, determine the position of the user profile information in the dialogue information, and obtain the position information of the position; The dialogue information, the location information, and the information type are respectively subjected to vector embedding processing to obtain dialogue vector, location vector, and type vector; The dialogue vector and the location vector are concatenated to obtain a dialogue vector carrying location information. The dialogue vector carrying location information is encoded to obtain an encoded vector; The encoding vector and the type vector are concatenated to obtain the information extraction vector; The information extraction vector is densified, and logistic regression is performed on the densified information extraction vector to obtain the probability that the user profile information belongs to each candidate object. Based on the probability that the user profile information belongs to each candidate object, the belonging object of the user profile information is predicted. The belonging object is the person or thing to which the user profile information belongs. The candidate object is any person or thing in the dialogue information.

2. The method according to claim 1, characterized in that, The training process of the context-related text pair matching network includes: acquiring dialogue training text; the dialogue training text includes the current dialogue training text and historical dialogue training text, the historical dialogue training text includes positive training text related to the current dialogue training text and negative training text unrelated to the current dialogue training text; inputting the current dialogue training text, positive training text and negative training text into the context-related text pair matching network, training the context-related text pair matching network, and obtaining the trained context-related text pair matching network.

3. The method according to any one of claims 1 to 2, characterized in that, The method further includes: The processing object is matched with each preset text for similarity, and a target object matching the processing object is determined from each preset text. The processing object includes any one of the user profile information, the information type, and the belonging object, and the target object is any one of the target user profile information, the target information type, and the target belonging object.

4. The method according to claim 3, characterized in that, The step of performing similarity matching between the processed object and each preset text, and determining the target object that matches the processed object from each preset text, includes: The similarity between the processed object and each of the preset texts is calculated using a literal similarity function or a shallow semantic similarity function.

5. An information extraction method, characterized in that, The method includes: Send an information retrieval request to the server; The user profile information sent by the server, the information type of the user profile information, and the object to which the user profile information belongs are obtained; wherein, the information extraction method described in claim 1 is used to determine the object to which the user profile information belongs. The user profile information, the information type, and the belonging object are displayed in the user information display area.

6. An information extraction device, characterized in that, The device includes: The dialogue information acquisition module is used to acquire dialogue information between users and service objects; the dialogue information includes current dialogue information and associated dialogue information; the process of determining the dialogue information includes: acquiring the current dialogue information between users and service objects, and the historical dialogue information corresponding to the current dialogue information; The current dialogue information is vectorized to obtain the current dialogue vector; the historical dialogue information is vectorized to obtain the historical dialogue vector. The current dialogue vector and the historical dialogue vector are concatenated to obtain the concatenated dialogue vector; The concatenated dialogue vector is input into the trained context-related text pair matching network to obtain the semantic similarity between the current dialogue information and each sub-dialogue information in the historical dialogue information, and to filter out the related dialogue information related to the current dialogue information from the historical dialogue information; wherein, the context-related text pair matching network is a BiMPM network. The key information extraction module is used to extract user profile information and the information type of the user profile information from the dialogue information; based on the dialogue information and the user profile information, determine the position of the user profile information in the dialogue information, and obtain the position information of the position; The relationship attribution module is used to perform vector embedding processing on the dialogue information, the location information and the information type respectively to obtain dialogue vector, location vector and type vector; The dialogue vector and the location vector are concatenated to obtain a dialogue vector carrying location information. The dialogue vector carrying location information is encoded to obtain an encoded vector; The encoding vector and the type vector are concatenated to obtain the information extraction vector; The information extraction vector is densified, and logistic regression is performed on the densified information extraction vector to obtain the probability that the user profile information belongs to each candidate object. Based on the probability that the user profile information belongs to each candidate object, the object to which the user profile information belongs is predicted. The candidate object is any person or thing in the dialogue information.

7. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 5.

8. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 5.

Images