Information identification method and device, and electronic device

Abstract

The application discloses an information recognition method and device and electronic equipment. Target text information associated with a target object is acquired. The target text information is recognized to obtain initial state information matched with the target object. An attribute label of the target object is acquired. The initial state information is calibrated based on the attribute label to obtain target state information of the target object. The calibration of the recognition result of the text information based on the attribute label of the object is realized, the recognition result is more accurate, and the experience effect of a user is improved.

Description

Technical Field

[0001] This application relates to the field of information processing technology, and more specifically to an information identification method, apparatus and electronic device. Background Technology

[0002] With the development of artificial intelligence, intelligent customer service and chatbots have emerged in intelligent application scenarios. Taking intelligent customer service systems as an example, they need to recognize the text information entered by users to ensure that the system can accurately understand the user's intent and complete the corresponding output.

[0003] However, user-input text information is diverse, and using fixed text semantic analysis cannot accurately grasp the user's intent or emotion, reducing the accuracy of text recognition. Consequently, the results or actions performed by intelligent customer service may fail to meet the user's needs, thus degrading the user experience. Summary of the Invention

[0004] In view of the above, this application provides the following technical solution:

[0005] An information identification method, comprising:

[0006] Retrieve the target text information associated with the target object;

[0007] The target text information is identified to obtain initial state information that matches the target object;

[0008] Obtain the attribute tags of the target object;

[0009] The initial state information is calibrated based on the attribute tags to obtain the target state information of the target object.

[0010] Optionally, obtaining the target text information associated with the target object includes:

[0011] If the text information input to the target object meets the input position condition, the text information is determined as the target text information;

[0012] or;

[0013] If the feedback information corresponding to the generated text information that matches the target object does not meet the target conditions, the text information is determined as the target text information.

[0014] Optionally, obtaining the attribute tags of the target object includes:

[0015] Acquire analytical auxiliary data that matches the target object;

[0016] Based on the analytical auxiliary data, the attribute labels of the target object are determined.

[0017] Optionally, calibrating the initial state information based on the apartment type label to obtain the target state information of the target object includes:

[0018] Detect whether the initial state information matches the state of the target object corresponding to the attribute label;

[0019] If not, the initial state information is calibrated based on the attribute labels to obtain the state information of the target object.

[0020] Optionally, calibrating the initial state information based on the attribute tags to obtain the target state information of the target object includes:

[0021] The attribute labels and the initial state information are input into the target recognition model to obtain the target state information of the target object. The target recognition model is a neural network model trained based on training samples. The training samples include the state information of the target object obtained based on text recognition, the attribute labels of the target object, and the labeled target state information of the target object.

[0022] Optionally, the method further includes:

[0023] Based on the target state information, the attribute labels of the target object are updated to obtain the updated attribute labels.

[0024] Optionally, the method further includes:

[0025] Determine the processing parameters that match the target state information;

[0026] The target text information is processed based on the processing parameters to obtain the processing result.

[0027] Optionally, determining the processing parameters that match the target state information includes:

[0028] Obtain the initial processing parameters corresponding to the target object, wherein the initial processing parameters are processing parameters that match the attribute tags of the target object;

[0029] Based on the target state information and the initial processing parameters, calculations are performed to obtain processing parameters that match the target state information.

[0030] An information identification device, comprising:

[0031] The first acquisition unit is used to acquire target text information associated with the target object;

[0032] The recognition unit is used to recognize the target text information and obtain initial state information that matches the target object;

[0033] The second acquisition unit is used to acquire the attribute tags of the target object;

[0034] A calibration unit is used to calibrate the initial state information based on the attribute labels to obtain the target state information of the target object.

[0035] An electronic device, comprising:

[0036] Memory, used to store applications and the data generated by the running of the applications;

[0037] A processor for executing the application to achieve:

[0038] Retrieve the target text information associated with the target object;

[0039] The target text information is identified to obtain initial state information that matches the target object;

[0040] Obtain the attribute tags of the target object;

[0041] The initial state information is calibrated based on the attribute tags to obtain the target state information of the target object.

[0042] As can be seen from the above technical solutions, the information recognition method, apparatus, and electronic device disclosed in this application acquire target text information associated with a target object; recognize the target text information to obtain initial state information matching the target object; acquire the attribute tags of the target object; and calibrate the initial state information based on the attribute tags to obtain the target state information of the target object. This achieves calibration of the recognition results of text information based on the object's attribute tags, making the recognition results more accurate and improving the user experience. Attached Figure Description

[0043] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only embodiments of this application. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0044] Figure 1 A flowchart illustrating an information identification method provided in Embodiment 1 of this application;

[0045] Figure 2 A schematic diagram illustrating the training of a sentiment prediction model provided in an embodiment of this application;

[0046] Figure 3 A schematic diagram illustrating an application scenario provided in an embodiment of this application;

[0047] Figure 4 This is a schematic diagram of the structure of an information identification device provided in Embodiment 2 of this application;

[0048] Figure 5 This is a schematic diagram of the structure of an electronic device provided in Embodiment 3 of this application. Detailed Implementation

[0049] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0050] See Figure 1 This is a flowchart illustrating an information recognition method provided in Embodiment 1 of this application. This method can be applied to scenarios involving the recognition of the state of a target object, facilitating subsequent processing based on the recognition results. Specifically, the method may include the following steps:

[0051] S101. Obtain the target text information associated with the target object.

[0052] Target text information refers to text information that meets specific conditions and is obtained in a real-world application scenario. It can be text information input by the target object, or text information converted from audio or other data generated by the target object. In practical applications, target text information can be inquiry text or message text input by the target object.

[0053] S102. Recognize the target text information to obtain initial state information that matches the target object.

[0054] After obtaining the target text information, it needs to be identified. This can be done through semantic recognition to obtain the corresponding state information, which can represent the user's current state, such as emotional state or response request state. Alternatively, the target text information can be output to a corresponding recognition model, which identifies the text information, obtains text features, and analyzes these features to obtain the initial state information of the target object. It should be noted that the initial state information in this embodiment is obtained solely based on the identification of the target text information.

[0055] S103. Obtain the attribute tags of the target object.

[0056] Among them, the attribute tags of the target object are generated based on the user's attribute information. This attribute information includes basic attribute information related to user characteristics, such as age, gender, occupation, etc., as well as the user's text attribute information, that is, tags generated based on the user's historical text, such as consultation satisfaction tags generated based on the user's historical consultation text.

[0057] In one possible implementation, the corresponding attribute tags can be obtained by creating user profiles of the target object. User profiles, also known as user roles, serve as an effective tool for outlining the target object and connecting its needs with design direction. A user profile is a visual representation of data associated with the target object, essentially achieving user information tagging. User profiles can be generated from the user's basic characteristic information and relevant information generated in the user's history. This historical information includes the user's browsing history, input text, and features related to the user's interests. Alternatively, a pre-trained user profile model can be used to obtain the target object's user profile. Correspondingly, the user's characteristic information can be converted into feature vectors and input into the user profile model, causing the model to output a user profile corresponding to the target object. After obtaining the target object's user profile, corresponding user tags can be obtained, such as statistical tags (including age, gender, region, etc.), interest tags (such as product preferences, interest types, etc.), and also mining tags (such as language feature tags, input text preference tags, etc.).

[0058] S104. Based on the attribute tags, the initial state information is calibrated to obtain the target state information of the target object.

[0059] In this embodiment, after obtaining the initial state information of the target object, the state in the initial state information is not taken as the final state of the target object. Instead, the initial state information is calibrated according to the attribute tags of the target object to obtain the target state information, and the state in the target state information is taken as the user's final state. When the target text is positive, the initial state of the target object obtained by recognizing the target text is also positive. However, when the user attribute tags obtained are negative, the initial state information is corrected by the attribute tags, and the final target state information can represent that the user is in a negative emotional state. For example, in the application scenario of service feedback, the target text entered by the user is "Your company's service attitude is really good, and the feedback is really positive." If the target text is recognized alone, it will show that the user is in a praising mood. However, after obtaining the user's attribute tags, the user's attribute tags include dissatisfaction with after-sales service, multiple inquiries, etc., which indicates that the target text entered by the user has a negative mood, and its essence is dissatisfaction with after-sales service and service feedback. Therefore, after calibrating the initial state information based on this attribute label, the target state information obtained is that the user is very dissatisfied with the current consultation results and feedback progress, and the after-sales processing progress needs to be accelerated.

[0060] Taking an AI-powered after-sales service system for a target product as an example, if a user encounters a problem or cannot use the product correctly after purchasing it, they can leave feedback online. The AI-powered after-sales service system will analyze the time and nature of the message to determine when to reply and generate a corresponding response, which will be sent to the user via their provided contact information. For example, a user's message on May 25th regarding a product issue reads, "When will you reply? This is really good after-sales service; the response is so prompt." A simple analysis of this text might suggest that the user wanted a reply and praised the after-sales service. Under normal processing conditions, since the user left the message on May 25th and the current time is May 27th, based on the processing conditions of time sequence and the severity of the issue, the current user status may be considered satisfactory and does not require immediate processing. That is, the processing priority coefficient of this user's message may be set to 3 (priority coefficient is 1-5, with a higher priority coefficient indicating higher priority processing). In other words, this message may be processed only after the messages from May 23rd and May 24th have been processed.

[0061] In this embodiment, after receiving the user's message, the priority parameter is not directly output based on text analysis. Instead, the priority parameter obtained from the message content is calibrated based on the user's attribute tags to determine the user's true status. If the obtained user attribute tags indicate frequent after-sales messages within a certain period, and the message content includes image information, text information, and tags related to negative feedback on recent replies, it can be determined that the user is dissatisfied with the current response and is passively resisting it. Therefore, the priority coefficient of this message is increased, for example, by setting it to 5. This allows the user to receive a reply before earlier messages with lower priority coefficients, and further transfers the user to human after-sales service to improve their satisfaction with the after-sales service.

[0062] As can be seen from the above technical solution, Embodiment 1 of this application provides an information recognition method that obtains target text information associated with a target object; identifies the target text information to obtain initial state information matching the target object; obtains the attribute tags of the target object; and calibrates the initial state information based on the attribute tags to obtain the target state information of the target object. This achieves calibration of the recognition results of text information based on the object's attribute tags, making the recognition results more accurate. This allows for the determination of the matching processing mode based on the actual state of the target object in subsequent processing, making the processing process more in line with the user's actual needs and thus improving the user experience.

[0063] The information recognition method in this application can be used to process all user-generated text to obtain the recognition result that most accurately reflects the target object's state. However, recognizing every piece of user-input text would consume excessive processing resources. Therefore, in this application, the abstract focuses on recognizing text information that meets the target conditions.

[0064] In one implementation, acquiring target text information associated with a target object includes: if the text information input by the target object satisfies an input location condition, identifying the text information as target text information. In a customer service system, the target object is the user making a consultation, typically through dialogue between the user and the intelligent customer service. During the target consultation period, the intelligent customer service responds to the user's input consultation information. The user can continue their consultation based on the intelligent customer service's response, but when the user stops inputting consultation information, it may indicate the consultation has ended, or the user is dissatisfied with the information output by the intelligent customer service. In this case, text information input by the target object that satisfies the input location condition is collected as target text information for identification. The input location condition can be set according to specific application scenarios. For example, in a customer service system, it could be the last text information output by the target object during their dialogue with the intelligent customer service. Alternatively, in a feedback scenario, it could be the text information corresponding to the target message location, such as the text information in the corresponding target feedback field. Specifically, for example, the target feedback field could be "Please provide feedback information regarding the new product" or "Please provide your opinion on this consultation."

[0065] In another implementation, obtaining target text information associated with the target object includes: in response to the generated text information matching the target object and corresponding feedback information not meeting the target conditions, determining the text information as target text information. Here, the target conditions characterize the conditions under which the target object is dissatisfied with the feedback information. When the text information input by the user is consultation text, the user will receive feedback information from the customer service system after inputting the text information. The user can also evaluate the feedback information, including whether they are satisfied or dissatisfied. When the user selects dissatisfaction with the feedback information or does not evaluate the feedback information, it can be considered that the feedback information output by the customer service system does not match the user's actual needs. Therefore, the user's text information will be identified to obtain status information that represents the user's true needs.

[0066] After identifying the target text information to obtain the user's initial state information, it is necessary to calibrate the initial state information based on the target object's attribute tags. Obtaining the target object's attribute tags includes: acquiring analytical auxiliary data matching the target object, and determining the target object's attribute tags based on the analytical auxiliary data. Analytical auxiliary data is all available data related to the target object, mainly including the target object's attribute data and text information with the same attributes as the target text information. The target object's attribute information is data at the target object profile level, such as the target object's gender, age, and occupation. Text information with the same attributes as the target text information includes data belonging to the same business type as the target text information. For example, if the target text information is a consultation message, the obtained analytical auxiliary information would be the user's historical text information before inputting the target text information. Specifically, this could include complaint information, consultation information, and product after-sales message information from within the past month. Therefore, the user's attribute tags can be updated using the user's corresponding historical text information, or the user's attribute tags can be further matched with corresponding tag importance parameters. For example, if a user's product review is positive, the corresponding product review tag will be "positive user." If the user adds a complaint within the past month, such as regarding untimely after-sales guidance, the user's product review tag will be updated based on the added comment. This means the tag for "positive user" can be changed to "dissatisfied with after-sales service." This ensures that user attribute tags reflect the user's true state, providing more accurate data for calibrating user status.

[0067] After obtaining the auxiliary analysis data, attribute tags for the target object can be generated based on the auxiliary analysis data. Specifically, when generating attribute tags, keywords can be extracted from the auxiliary analysis data, and the extracted keywords can be merged to obtain the corresponding tag information. This application does not restrict the specific method of generating attribute tags.

[0068] In one implementation, the step of calibrating the initial state information based on attribute tags to obtain the target state information of the target object includes:

[0069] Check whether the initial state information matches the state of the target object corresponding to the attribute label. If not, calibrate the initial state information based on the attribute label to obtain the state information of the target object.

[0070] Attribute tags can characterize the features of a target object. If the initial state information represents the user's satisfaction with the current customer service system, and the attribute tags represent the user's satisfaction with previous after-sales consultation information, then the initial state information matches the attribute tags, and calibration of the initial state information is unnecessary. Conversely, if there is a mismatch or a low degree of matching, the attribute tags will be used to calibrate the initial state information.

[0071] In one implementation, the step of calibrating the initial state information based on attribute tags to obtain the target state information of the target object includes:

[0072] Input the attribute labels and initial state information into the target recognition model to obtain the target state information of the target object.

[0073] The target recognition model is a neural network model trained based on training samples, which include the state information of the target object obtained from text recognition, the attribute labels of the target object, and the target state information of the labeled target object.

[0074] Specifically, during the model training phase, it is necessary to first obtain historical text information of multiple target objects and their corresponding attribute labels as a training sample set. Each sample in the training sample set needs to be manually labeled, accurately indicating the user's actual state for each sample. After labeling the target objects' states in the training data, the target recognition model can be trained based on the labeled training data. When training the model, the recognition model can employ information classification models, such as SVM (Support Vector Machine), CNN (Convolutional Neural Network), or LSTM (Long Short-Term Memory); or information extraction models, such as CRF (Conditional Random Field algorithm), LSTM+CRF, etc., or a combination of text classification and information extraction models can be effectively integrated.

[0075] After completing the training of the target recognition model, the initial state information and attribute labels identified based on the target text information can be input into the target recognition model. The model will then identify and process the above information to obtain the target state information.

[0076] It should be noted that the above target recognition model is one implementation method based on model processing. The corresponding model can also have other structures, and the corresponding training samples will differ depending on the model structure. For example, historical target text information, attribute labels, and target state information can be used as annotation information to generate training samples. Then, the model can be trained based on these training samples to obtain the target model. This way, the input data to the target model consists of target text information and attribute labels, and the output information of the target model is target state information.

[0077] To enable the attribute labels of a target object to represent its features in real time, the attribute labels can be updated based on the target's state information. This updated attribute labeling allows subsequent identification to utilize the updated labeling, resulting in more accurate state recognition of the target object.

[0078] For example, if the initial state information indicates that the user's output text is positive, and the attribute label is a user satisfaction level of 3 with the intelligent customer service (e.g., a satisfaction score range of 0-5, with lower scores indicating lower satisfaction), then the target state information indicates that the user's output text is feedback of dissatisfaction with the current intelligent customer service. Then, based on the target state information, the updated attribute label is a user satisfaction level of 2 with the intelligent customer service.

[0079] In one embodiment of this application, based on the foregoing embodiments, the method further includes:

[0080] Determine the processing parameters that match the target state information;

[0081] The target text information is processed based on the processing parameters to obtain the processing result.

[0082] Among them, the processing parameters represent the level of processing of the target text information, which can be priority parameters. If the target state information represents the emotional state level of the target object, and the emotional state level is divided into 1-5 levels, the lower the level, the more dissatisfied the object is with the current consultation result. When the emotional state level corresponding to the target state information is identified as 2, it can be known that the target object is very dissatisfied with the current consultation result. Therefore, the corresponding processing parameter has a higher priority, so that the consultation problem reported by the target object is processed first to improve the target object's satisfaction.

[0083] Specifically, determining the processing parameters matching the target state information includes: obtaining the initial processing parameters corresponding to the target object; and calculating the processing parameters matching the target state information based on the target state information and the initial processing parameters. The initial processing parameters are the processing parameters matching the attribute tags of the target object. Different attribute tags can match corresponding processing parameters, and different state information corresponds to different weight values. The processing parameters matching the target state information can be calculated based on the weight values ​​and the initial processing parameters.

[0084] Taking a real-world application scenario of an intelligent customer service system as an example, the text information includes user feedback or suggestions regarding products or services. The user's state includes their emotional state, with attribute tags comprising basic and business attribute tags. Basic attribute tags include demographic characteristics such as age, gender, and occupation, while business attribute tags include user-reported repairs within the past 30 days, malfunctions of newly purchased devices within the past 3 months, and historical complaints. By inputting this text information into a text sentiment recognition model and obtaining a user sentiment score, a sample vector is constructed based on the user's historical tag data. This vector is then manually labeled, allowing for the training of a sentiment prediction model that integrates user tags. This approach aims to deeply mine user emotions from dimensions beyond the text itself.

[0085] Specifically, when generating this sentiment prediction model, each training sample in the training sample set includes text information, user attribute tags, and labeled user sentiment. The model is trained using this sample set until the predicted user sentiment output by the trained model is consistent with or similar to the labeled user sentiment. At this point, the training and optimization process stops, resulting in the sentiment prediction model. This allows subsequent input of the user's corresponding text information and attribute tags into the sentiment prediction model to obtain the corresponding sentiment for that user. The sentiment can be positive or negative. If the labeled user sentiment type in the training samples is more specific, the sentiment type output by the sentiment prediction model can also be more specific. For example, the output sentiment prediction information can include corresponding level information, specifically, it can include level one positive sentiment, level two positive sentiment, level one negative sentiment, level two negative sentiment, etc.

[0086] For example, if a user inputs the text "Your service is excellent, I hope you always are this good," and the corresponding attribute tags are "female, new user, no product reviews," then inputting this information into the sentiment prediction model will result in a positive sentiment. As another example, if the user inputs the same text, but the corresponding attribute tags are "female, age 30-40, long-time user, 5 past complaints, 3 after-sales service requests, high rate of negative after-sales service reviews," then inputting this information into the sentiment prediction model will result in a negative sentiment, meaning the user is not satisfied with the current service—it's a case of saying the opposite of what they mean.

[0087] Based on the sentiment prediction model described above, see [link / reference]. Figure 2 This is a training diagram of a sentiment prediction model provided in an embodiment of this application. The sentiment prediction model is a deep learning model, which is generally composed of various model layers, such as... Figure 2 As shown, the model layers of this sentiment prediction model include X layers, which can be categorized into convolutional layers, pooling layers, and fully connected layers. It should be noted that the specific structure can vary depending on the choice of neural network architecture. Figure 2 The model structure is represented by Layer1, Layer2, ..., Layer X. The model comprises two branches: input information includes text (e.g., "I wish you all a happy birthday") and user history tags (e.g., "Purchased a device that is currently damaged and has been repeatedly reported for repair"). These are then fed into corresponding extraction networks to extract text and tag features. The extracted features are then fed into sentence embedding and user tag embedding networks, further entering the model layer to obtain the predicted sentiment score Y. i And based on the target sentiment score y i The model is trained to make the final predicted emotion score close to the target emotion score. During training, the model parameters are continuously adjusted to ensure that the predicted emotion score approaches the target emotion score, such as adjusting the relevant parameters in the model layers and extraction network.

[0088] Specifically, the training samples consist of the user's corresponding text, attribute labels, and labeled user sentiment scores, where the labeled user sentiment scores are the target sentiment scores. The input feature sequence is obtained by constructing the text, attribute labels, and target sentiment scores from the corresponding sample information. Figure 2Taking a training sample as an example, the text "I wish you all the best in your business" is input into the extraction network as a text feature, and "Purchased within a month and repeatedly reported for repair" is input into the corresponding extraction network as a user attribute label. The extraction network extracts the corresponding relevance prediction features, and then the semantic features are input into the corresponding model layer, such as the fully connected network layer. The predicted relevance between the user sentiment score labeled in the training sample and the text and attribute label is obtained from the fully connected network layer. Then, the corresponding loss function value can be determined based on the predicted relevance and the actual relevance. Based on the loss function value, the weight parameters of each model layer are adjusted so that the final predicted sentiment score Yi is the same as or close to the target sentiment score yi.

[0089] It should be noted that the focus of constructing the sentiment prediction model in this embodiment is the selection of sample information. This includes not only the text information associated with the target object but also attribute tags matching the target object. These attribute tags can be selected based on different application scenarios or sentiment prediction needs. For example, in a sentiment recognition scenario where users evaluate products, user attribute tags can be selected based on the attribute tags corresponding to the product. Specifically, these could include user evaluation tags, interest product tags, evaluation text clustering tags, etc. Similarly, in a sentiment recognition scenario where users provide feedback on new products, user attribute tags could include user feature tags (such as whether the user is a new user), user information, user interest tags for waiting for new products, user usage time tags, etc. Therefore, in this embodiment, the model training process does not solely rely on text information as training samples; it also incorporates the user-associated attribute tags corresponding to the text information as part of the training samples, ensuring that the output user state (such as user sentiment) is more closely matched to the user and the current scenario.

[0090] Based on the aforementioned sentiment prediction model, if the user inputs text such as "I wish your business will get better and better," it is identified as a positive evaluation through text recognition. In this embodiment, by incorporating the user's historical tag data and training the model with both, the result can be corrected to a negative evaluation. This allows for the identification of the user's actual emotions in statements such as "sarcasm."

[0091] Traditional user sentiment recognition typically performs semantic understanding and modeling at the plain text level. This involves first representing each sentence spoken by the user as a text feature, then obtaining token embeddings or sentence embeddings using algorithms such as Word2Vec, ELMo, and BERT. These are then used to train a neural network to predict sentiment polarity and sentiment scores. In this embodiment, in addition to text features, historical attribute tags are added, such as historical service tags, for example, repeat repairs (repeated repairs within 1 month, 3 months, 6 months), new purchase device malfunctions (new purchase device malfunctions within 1 month, 3 months, 6 months), historical negative reviews (negative reviews within 1 month, 3 months, 6 months), historical complaints (negative reviews within 1 month, 3 months, 6 months), and demographic features such as age, gender, and customer occupation. These features are then vectorized using Extractor Net (e.g., ...). Figure 2 The user tag embedding network is fed into the neural network for training, supplementing feature information beyond the text and improving the accuracy of personalized user sentiment analysis.

[0092] For example, see Figure 3 This is a schematic diagram illustrating an application scenario provided by an embodiment of this application. In the application scenario of intelligent customer service, the user attribute tags of online service access user A include female, 30-40 years old, journalist, newly purchased device with malfunction within 3 months, complaint within 1 month, and a sentiment score of 0.83 for the last consultation (score range 0-5, the lower the score, the worse the sentiment).

[0093] After a brief explanation of the problem, User A lost patience with the intelligent customer service and had no intention of continuing the conversation. User A's final input was, "Your service is fantastic, I wish you continued success." This input text did not exhibit obvious negative emotion. When using text-level features to build a sentiment analysis model, it was difficult to detect this negative emotion even with contextual information. Adjusting the training sample set had little effect, and the final model prediction was positive. However, through the embodiments of this application, in addition to text features, the model was trained by combining user historical service tags and user attribute tags, ultimately identifying the user's clearly negative emotion, thus improving the accuracy of sentiment identification in this situation. Furthermore, when this emotional state is identified, the user's inquiry request can be prioritized, reducing the further escalation of negative emotions and improving the user's experience with the intelligent customer service system.

[0094] See Figure 4This is a schematic diagram of the structure of an information recognition device provided in Embodiment 2 of this application. The technical solution in this embodiment is mainly used to improve the accuracy of user status recognition.

[0095] Specifically, the apparatus in this embodiment may include the following units:

[0096] The first acquisition unit 401 is used to acquire target text information associated with the target object;

[0097] The recognition unit 402 is used to recognize the target text information to obtain initial state information that matches the target object;

[0098] The second acquisition unit 403 is used to acquire the attribute tags of the target object;

[0099] The calibration unit 404 is used to calibrate the initial state information based on the attribute label to obtain the target state information of the target object.

[0100] As can be seen from the above technical solution, Embodiment 2 of this application discloses an information recognition device that acquires target text information associated with a target object; recognizes the target text information to obtain initial state information matching the target object; acquires the attribute tags of the target object; and calibrates the initial state information based on the attribute tags to obtain the target state information of the target object. This achieves calibration of the recognition results of text information based on the object's attribute tags, making the recognition results more accurate and improving the user experience.

[0101] In one implementation, the first acquisition unit 401 includes:

[0102] The first determining subunit is used to determine the text information as target text information if the text information input by the target object satisfies the input position condition;

[0103] or;

[0104] The second determining subunit is used to determine the text information as target text information in response to feedback information corresponding to the generated text information matching the target object not meeting the target conditions.

[0105] Optionally, the second acquisition unit 403 includes:

[0106] The first acquisition subunit is used to acquire analytical auxiliary data that matches the target object;

[0107] The third determining subunit is used to determine the attribute labels of the target object based on the analysis auxiliary data.

[0108] In one embodiment, the calibration unit 404 includes:

[0109] A detection subunit is used to detect whether the initial state information matches the state of the target object corresponding to the attribute label;

[0110] The calibration subunit is used to calibrate the initial state information based on the attribute label, if not, to obtain the state information of the target object.

[0111] Optionally, the calibration unit 404 is specifically configured to include:

[0112] The attribute labels and the initial state information are input into the target recognition model to obtain the target state information of the target object. The target recognition model is a neural network model trained based on training samples. The training samples include the state information of the target object obtained based on text recognition, the attribute labels of the target object, and the labeled target state information of the target object.

[0113] Optionally, the device further includes:

[0114] The update unit is used to update the attribute labels of the target object based on the target state information to obtain the updated attribute labels.

[0115] Furthermore, the device also includes:

[0116] A parameter determination unit is used to determine processing parameters that match the target state information;

[0117] The processing unit is used to process the target text information based on the processing parameters to obtain the processing result.

[0118] Optionally, the parameter determination unit is specifically used for:

[0119] Obtain the initial processing parameters corresponding to the target object, wherein the initial processing parameters are processing parameters that match the attribute tags of the target object;

[0120] Based on the target state information and the initial processing parameters, calculations are performed to obtain processing parameters that match the target state information.

[0121] It should be noted that the details of each unit and its specific implementation in this embodiment can be found in the corresponding content above, and will not be described in detail here.

[0122] See Figure 5 This is a schematic diagram of the structure of an electronic device provided in Embodiment 3 of this application. The technical solution of this embodiment is mainly used to improve the accuracy of user status recognition.

[0123] Specifically, the electronic device in this embodiment may include the following structure:

[0124] Memory 501 is used to store the application and the data generated by the application during its operation;

[0125] Processor 502 is configured to execute the application to achieve:

[0126] Retrieve the target text information associated with the target object;

[0127] The target text information is identified to obtain initial state information that matches the target object;

[0128] Obtain the attribute tags of the target object;

[0129] The initial state information is calibrated based on the attribute tags to obtain the target state information of the target object.

[0130] As can be seen from the above technical solutions, Embodiment 3 of this application discloses an electronic device that acquires target text information associated with a target object; identifies the target text information to obtain initial state information matching the target object; acquires the attribute tags of the target object; and calibrates the initial state information based on the attribute tags to obtain the target state information of the target object. This achieves calibration of the recognition results of text information based on the object's attribute tags, making the recognition results more accurate and improving the user experience.

[0131] It should be noted that the specific implementation of the processor in this embodiment can be referred to the corresponding content above, and will not be described in detail here.

[0132] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatus disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple; relevant parts can be referred to the method section.

[0133] Those skilled in the art will further recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0134] The steps of the methods or algorithms described in conjunction with the embodiments disclosed herein can be implemented directly by hardware, a software module executed by a processor, or a combination of both. The software module can be located in random access memory (RAM), main memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disk, removable disk, CD-ROM, or any other form of storage medium known in the art.

[0135] The above description of the disclosed embodiments enables those skilled in the art to make or use this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. An information identification method, comprising: Retrieve the target text information associated with the target object; The target text information is identified to obtain initial state information that matches the target object; Obtain the attribute tags of the target object; the attribute tags include tags generated based on text generated from user history, and the attribute tags include after-sales service information; generate a user profile through the basic feature information corresponding to the target object and the information generated from user history, and obtain the attribute tags corresponding to the target object based on the user profile; The initial state information is calibrated based on the attribute tags to obtain the target state information of the target object.

2. The method according to claim 1, wherein obtaining the target text information associated with the target object includes: If the text information input to the target object meets the input position condition, the text information is determined as the target text information; or; If the feedback information corresponding to the generated text information that matches the target object does not meet the target conditions, the text information is determined as the target text information.

3. The method according to claim 1, wherein obtaining the attribute tags of the target object includes: Acquire analytical auxiliary data that matches the target object; Based on the analytical auxiliary data, the attribute labels of the target object are determined.

4. The method according to claim 1, wherein calibrating the initial state information based on the attribute tags to obtain the target state information of the target object includes: Detect whether the initial state information matches the state of the target object corresponding to the attribute label; If not, the initial state information is calibrated based on the attribute labels to obtain the state information of the target object.

5. The method according to claim 1, wherein calibrating the initial state information based on the attribute tags to obtain the target state information of the target object includes: The attribute labels and the initial state information are input into the target recognition model to obtain the target state information of the target object. The target recognition model is a neural network model trained based on training samples. The training samples include the state information of the target object obtained based on text recognition, the attribute labels of the target object, and the labeled target state information of the target object.

6. The method according to claim 1, further comprising: Based on the target state information, the attribute labels of the target object are updated to obtain the updated attribute labels.

7. The method according to claim 1, further comprising: Determine the processing parameters that match the target state information; The target text information is processed based on the processing parameters to obtain the processing result.

8. The method according to claim 7, wherein determining the processing parameters matching the target state information includes: Obtain the initial processing parameters corresponding to the target object, wherein the initial processing parameters are processing parameters that match the attribute tags of the target object; Based on the target state information and the initial processing parameters, calculations are performed to obtain processing parameters that match the target state information.

9. An information identification device, comprising: The first acquisition unit is used to acquire target text information associated with the target object; The recognition unit is used to recognize the target text information and obtain initial state information that matches the target object; The second acquisition unit is used to acquire the attribute tags of the target object; the attribute tags include tags generated based on text generated by the user's history, and the attribute tags include after-sales service information; a user profile is generated by the basic feature information corresponding to the target object and the information generated by the user's history, and the attribute tags corresponding to the target object are obtained based on the user profile. A calibration unit is used to calibrate the initial state information based on the attribute labels to obtain the target state information of the target object.

10. An electronic device, comprising: Memory, used to store applications and the data generated by the running of the applications; A processor for executing the application to achieve: Retrieve the target text information associated with the target object; The target text information is identified to obtain initial state information that matches the target object; Obtain the attribute tags of the target object; the attribute tags include tags generated based on text generated from user history, and the attribute tags include after-sales service information; generate a user profile through the basic feature information corresponding to the target object and the information generated from user history, and obtain the attribute tags corresponding to the target object based on the user profile; The initial state information is calibrated based on the attribute tags to obtain the target state information of the target object.

Images