Text input recommendation method and device, computer equipment and storage medium

Abstract

The invention relates to the technical field of artificial intelligence, and discloses a text input recommendation method and device, computer equipment and a storage medium in the fields of financial science and technology and medical health. According to the method, when a newly added input word is detected, context information corresponding to the newly added input word is obtained, and a target intention label is determined according to the context information; performing predictive analysis on the newly added input word, the context information and the target intention tag through a preset input recommendation model to obtain at least one initial recommendation word; filtering and sorting all the initial recommendation words to obtain a recommendation word sequence, and sending the recommendation word sequence to a client; and when a recommended word selection instruction corresponding to the recommended word sequence sent by the client is received, determining a target input word from the recommended word sequence according to the recommended word selection instruction. According to the method, the availability and accuracy of text input recommendation words are ensured, and the text input efficiency is improved.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and more particularly to a text input recommendation method, apparatus, computer device, and storage medium in the fields of fintech and healthcare. Background Technology

[0002] Text input is ubiquitous in the fintech and healthcare sectors. In the financial insurance industry, customer service representatives use text input to provide services such as business consultations, claims inquiries, and complaint handling. In the healthcare industry, medical professionals use text input to collect and integrate diverse data, including basic information about patients, past medical histories, and health monitoring data, providing a comprehensive data foundation for diagnosis and treatment.

[0003] In existing technologies, text input requires spelling out each word individually, resulting in very slow input speeds, especially when processing large amounts of information, thus reducing text input efficiency. Furthermore, manual spelling input is prone to errors and omissions, leading to disputes in insurance and healthcare services. While some input suggestion tools can be used, these tools can only make simple associations for single characters, and the content they cannot suggest is often unsuitable, and they are prone to errors involving synonyms, increasing the error rate of text input. Therefore, a new text input recommendation method is urgently needed. Summary of the Invention

[0004] Therefore, it is necessary to provide a text input recommendation method, apparatus, computer device, and storage medium to address the aforementioned technical problems, thereby solving the issues of low efficiency and susceptibility to errors in existing text input methods.

[0005] A text input recommendation method, comprising: When a new input word is detected, the context information corresponding to the new input word is obtained, and the target intent tag is determined based on the context information; The newly added input words, context information, and target intent tags are predicted and analyzed by a preset input recommendation model to obtain at least one initial recommendation word; All the initial recommended words are filtered and sorted to obtain a recommended word sequence, and the recommended word sequence is sent to the client; Upon receiving a recommendation word selection instruction sent by the client corresponding to the recommendation word sequence, the target input word is determined from the recommendation word sequence according to the recommendation word selection instruction.

[0006] A text input recommendation device, comprising: The information acquisition module is used to acquire context information corresponding to the newly added input word when a new input word is detected, and to determine the target intent tag based on the context information; The model prediction module is used to predict and analyze the newly added input words, context information and target intent tags through a preset input recommendation model to obtain at least one initial recommendation word. The filtering and sorting processing module is used to filter and sort all the initial recommended words to obtain a recommended word sequence, and send the recommended word sequence to the client; The recommended word selection module is used to determine the target input word from the recommended word sequence according to the recommended word selection instruction sent by the client when it receives the recommended word selection instruction corresponding to the recommended word sequence.

[0007] A computer device includes a memory, a processor, and computer-readable instructions stored in the memory and executable on the processor, wherein the processor implements the above-described text input recommendation method when executing the computer-readable instructions.

[0008] A computer-readable storage medium storing computer-readable instructions that, when executed by one or more processors, cause the one or more processors to perform the text input recommendation method described above.

[0009] In the aforementioned text input recommendation method, apparatus, computer device, and storage medium, the method obtains context information corresponding to the newly added input word upon detection and determines the target intent tag based on the context information; it then performs predictive analysis on the newly added input word, context information, and target intent tag using a preset input recommendation model to obtain at least one initial recommended word; it filters and sorts all initial recommended words to obtain a recommended word sequence and sends the sequence to the client; upon receiving a recommended word selection instruction corresponding to the recommended word sequence from the client, it determines the target input word from the recommended word sequence according to the instruction. This invention, based on the newly added input word, also combines context information and target intent tags to automatically predict and generate the recommended word sequence that the user needs to input next, ensuring the usability and accuracy of text input recommendations. Simultaneously, it facilitates users in quickly selecting the target input word from the recommended word sequence as needed, accelerating text input speed and improving text input efficiency. Attached Figure Description

[0010] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments of the present invention will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0011] Figure 1 This is a schematic diagram of an application environment for a text input recommendation method according to an embodiment of the present invention; Figure 2 This is a flowchart illustrating a text input recommendation method according to an embodiment of the present invention; Figure 3 This is a schematic diagram of a text input recommendation device in one embodiment of the present invention; Figure 4 This is a schematic diagram of a computer device according to an embodiment of the present invention. Detailed Implementation

[0012] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0013] The text input recommendation method provided in this embodiment can be applied to, for example, Figure 1 In this application environment, the client and server communicate. The client includes, but is not limited to, various personal computers, laptops, smartphones, tablets, and portable wearable devices. The server can be implemented using a standalone server or a server cluster consisting of multiple servers. The client obtains the user's pre-spelled input words and sends them to the server. After obtaining the input words, the server performs predictive analysis to generate recommended words, which are then sent to the client for the user to directly select as their next input word, eliminating the hassle of spelling.

[0014] The text input recommendation method described in this embodiment can be applied to the financial and insurance industry. In insurance customer service scenarios, customer service personnel provide services such as business consultation, claims inquiry, and complaint handling to insurance customers through text input. By using the recommended words from the text input recommendation method, customer service personnel can quickly respond to customer inquiries, reduce text input errors caused by manual spelling, improve service quality, and enhance customer satisfaction.

[0015] The text input recommendation method described in this embodiment can also be applied to the healthcare industry. Healthcare professionals need to collect and integrate diverse data such as basic patient information, past medical history, and health monitoring data through text input to provide a comprehensive data foundation for diagnosis and treatment. Healthcare professionals face complex and varied patient situations daily; untimely text input can lead to long waiting times, affecting patient access to care, and is also prone to spelling errors or omissions. Using recommended words from the text input recommendation method significantly speeds up text input, and the recommended words predicted based on contextual information can reduce spelling errors or omissions.

[0016] In one embodiment, such as Figure 2 As shown, a text input recommendation method is provided, including the following steps S10-S40: S10. When a new input word is detected, obtain the context information corresponding to the new input word, and determine the target intent tag based on the context information.

[0017] Understandably, when text input begins, the user enters spelling information in the input box on the client side. The client then uses the words the user has already spelled as new input words and sends them to the server. New input words refer to newly added characters in the text input box, specifically the words or phrases generated by the user's latest input operation.

[0018] When the server detects a new input word, it triggers a text input recommendation process to suggest the next input word. The server obtains the context information corresponding to the new input word and determines the target intent tag based on this information. Context information refers to the text composed of all input words preceding the new input word in a continuous text input scenario; for example, all conversation content from the start of the interaction to the point before the new input word appears. The target intent tag represents the type of behavioral purpose expressed by the user's input context information. For example, if the context information is "What's the weather like tomorrow?", then the predefined target intent tag "Check the weather" can be matched.

[0019] S20. The newly added input words, context information and target intent tags are predicted and analyzed using a preset input recommendation model to obtain at least one initial recommended word.

[0020] Understandably, the server inputs the new input word, context information, and target intent label into a pre-set input recommendation model for predictive analysis, resulting in one or more initial recommended words. The pre-set input recommendation model is a pre-trained neural network model that predicts the next possible input word based on the latest input word, context information, and the intent label corresponding to that context information. The pre-set input recommendation model is a Recurrent Neural Network (RNN), a neural network model specifically designed for processing sequential data (such as text, speech, time series, etc.). Specifically, the pre-set input recommendation model can be a Long Short-Term Memory (LSTM) model, or it can be a Gated Recurrent Unit (GRU). The initial recommended words are input words that may meet the user's needs in the next moment, obtained based on the prediction results of the pre-set input recommendation model.

[0021] S30. Filter and sort all the initial recommended words to obtain a recommended word sequence, and send the recommended word sequence to the client.

[0022] Understandably, after obtaining initial recommended words through a preset input recommendation model, the server does not directly display these initial recommended words. Instead, it filters and sorts all the initial recommended words and sends the filtered and sorted recommended word sequence to the client for display to the user. The filtering and sorting process involves removing words from the initial recommended words that do not meet grammatical requirements and reordering them according to the reasonableness of their combination with newly added input words (the latest input words that have completed the input operation). The recommended word sequence refers to a set of recommended words obtained by sorting the remaining initial recommended words after filtering.

[0023] S40. Upon receiving a recommendation word selection instruction sent by the client corresponding to the recommendation word sequence, determine the target input word from the recommendation word sequence according to the recommendation word selection instruction.

[0024] Understandably, after filtering and sorting, the recommended word sequence may contain only one remaining initial recommended word, or it may contain multiple remaining initial recommended words. Users can select one word from the recommended word sequence as needed. The client generates a recommended word selection instruction based on the user's selection and sends this instruction to the server. The recommended word selection instruction refers to selecting a word from the recommended word sequence as the selection signal for the actual word entered in the input box at the next moment.

[0025] After sending the recommended word sequence to the client, the server displays it to the user in the input box and determines whether it has received a recommended word selection instruction corresponding to the recommended word sequence from the client within a preset waiting time. The preset waiting time is a pre-defined maximum time threshold for waiting for the user to actively select an input word. If the server receives a recommended word selection instruction within the preset waiting time, it determines the target input word from the recommended word sequence based on the instruction. The target input word is the actual input word following the newly added input word. Alternatively, if the server does not receive a recommended word selection instruction within the preset waiting time but detects a user input instruction, it determines the target input word based on the user input instruction. The user input instruction refers to the operation instruction where the user manually spells out the word. If the server neither receives a recommended word selection instruction nor detects a user input instruction within the preset waiting time, it determines the first word in the recommended word sequence as the target input word.

[0026] Furthermore, after the server determines the target input word, it sends the target input word to the client's input box, completing the input operation of the target input word. At this time, the target input word becomes the new input word for the next moment (the latest input word that has completed the input operation). When the server detects a new input word, it triggers a text input recommendation method to recommend the input word for the moment after that, until the input is completed or the session interaction ends.

[0027] This embodiment, upon detecting a new input word, acquires the context information corresponding to the new input word and determines the target intent tag based on the context information. It then uses a preset input recommendation model to predict and analyze the new input word, context information, and target intent tag to obtain at least one initial recommended word. All initial recommended words are filtered and sorted to obtain a recommended word sequence, which is then sent to the client. Upon receiving a recommended word selection instruction from the client corresponding to the recommended word sequence, the target input word is determined from the recommended word sequence according to the instruction. This embodiment, based on the new input word, also combines context information and target intent tags to automatically predict and generate the recommended word sequence that the user will need to input next, ensuring the usability and accuracy of text input recommendations. Simultaneously, it allows users to quickly select the target input word from the recommended word sequence as needed, accelerating text input speed and improving text input efficiency.

[0028] In one embodiment, step S10, namely determining the target intent tag based on the context information, includes: S101. Perform word segmentation on the context information to obtain multiple context words; S102. Find the segmentation entity tags of each context segmentation from the preset entity tag dictionary; S103. The target intent tag is obtained by predicting and analyzing all the segmented entity tags through a preset intent prediction model.

[0029] Understandably, when the server determines the target intent tag based on contextual information, it first performs word segmentation on the contextual information, resulting in multiple contextual words. Word segmentation refers to the process of dividing a continuous text sequence into grammatically meaningful word units. Word segmentation can provide structured input for subsequent text analysis (such as intent analysis and information retrieval). Contextual word segmentation refers to the grammatically meaningful words segmented from contextual information.

[0030] Next, the server retrieves the entity tags for each context word segmentation from the pre-defined entity tag dictionary. The pre-defined entity tag dictionary is a pre-built data table used to store entity tags associated with different words. This dictionary can be customized for specific domains (such as fintech or healthcare), allowing the entity tags to focus on key domain information and reduce noise interference. Entity tags are identifiers used to label the entity category to which each word segmentation unit belongs in the text; for example, in insurance claims, they might include the name, license plate number, insurance name, time, and location.

[0031] Finally, the server uses a pre-set intent prediction model to predict and analyze all segmented entity tags to obtain the target intent tag. The pre-trained intent prediction model is a neural network model used to predict the intent tag corresponding to combinations of multiple segmented entity tags. When the segmented entity tags are "name, license plate number, report an accident, time and location", the corresponding target intent tag is "car insurance claim".

[0032] This embodiment combines contextual word segmentation, entity tag lookup, and intent prediction to more efficiently and accurately determine the target intent tag. Specifically, by clarifying key information through entity tags, unstructured text is transformed into structured input, which helps the model more accurately match the intent during prediction, avoiding errors that may occur due to semantic ambiguity when directly using raw text to predict intent. Furthermore, the preset entity tag dictionary supports updates and expansions to meet the needs of a wide range of users.

[0033] In one embodiment, step S20, namely, predicting and analyzing the newly added input word, context information, and target intent tag using a preset input recommendation model to obtain at least one initial recommended word, includes: S201. The newly added input word, context information and target intent tag are predicted and analyzed by a preset input recommendation model to obtain at least one candidate recommendation word and the candidate recommendation probability value corresponding to each candidate recommendation word. S202. Select an initial recommended word of a preset recommended quantity from all the candidate recommended words according to the candidate recommended probability value.

[0034] Understandably, before the server performs predictive analysis through the preset input recommendation model, it is necessary to first complete the training of the preset input recommendation model. In a specific embodiment, a large amount of text dialogue data is collected from historical auto insurance service communications, including customer consultations, complaints, claims applications, and real-time online chat content, etc. Then, preprocessing is performed on these text dialogue data, and the preprocessing includes cleaning, word segmentation, stop word removal, and stemming. Among them, cleaning refers to the process of removing irrelevant information, such as removing advertisements, non-auto insurance-related topics, etc., and performing desensitization processing on sensitive information. Word segmentation refers to splitting the text into words or phrases for subsequent processing. Stop word removal refers to removing common stop words, such as "de", "shi", etc., to reduce noise. Stemming refers to restoring the words to their basic forms to reduce the diversity of vocabulary. Then, the preprocessed text is converted into a word vector sequence using word embedding techniques (such as Word2Vec, GloVe). Extract the historical context information of the dialogue (such as the content of the previous few rounds of dialogue), and generate intent labels according to the dialogue content, such as auto insurance consultation, complaint, claim, etc. Training sample data is obtained based on the word vector sequence and intent labels.

[0035] Next, the constructed recurrent neural network model is trained based on the training sample data to obtain a trained preset input recommendation model. Among them, the cross-entropy loss function (Cross-Entropy Loss) is used to measure the difference between the predicted input word and the actual input word of the model. The cross-entropy loss function is . Among them, represents the cross-entropy loss function value, represents the sequence length of the actual input word, represents the serial number of the actual input word, represents the serial number of the predicted input word, represents the total number of vocabulary data of the predicted input word, represents the th actual input word, represents that the th predicted input word of the model is the th actual input word with a probability. At the same time, an optimizer is used to update the model parameters, including but not limited to optimizers such as Adam and SGD.

[0036] After training the preset input recommendation model, a practical application is implemented. The server inputs new input words, context information, and target intent tags into the preset input recommendation model. First, the server obtains the target model parameters corresponding to the target intent tag and generates a target input recommendation model based on these parameters and the preset input recommendation model. Target model parameters refer to model parameters related to a specific intent tag, including model weights, bias terms, and hyperparameters, used for input word recommendation behavior in customized intent scenarios. The preset input recommendation model can be viewed as a general, predefined recommendation model structure. The target input recommendation model is an input recommendation model modified from the model parameters in the preset input recommendation model; it is specifically designed for handling specific intent recommendation tasks. Next, the target input recommendation model performs predictive analysis on the new input word and context information to obtain at least one candidate recommendation word and the corresponding candidate recommendation probability value. Specifically, the output probability distribution of the target input recommendation model is as follows: .in, Indicates the first The probability values ​​of candidate recommended words (predicted input words). Indicates the hidden layer state. This represents the weight matrix of the output layer. This represents the bias term of the output layer. This represents the normalized exponential function.

[0037] Finally, a predetermined number of initial recommended words are selected from all candidate recommended words based on their probability values. Specifically, the candidate recommended words can be sorted according to their probability values, and the top-ranked candidate recommended words are selected as the initial recommended words. The candidate recommendation probability value refers to the probability that the model predicts the word that might be input at the next moment. The predetermined number of recommended words is a pre-set maximum threshold value used to limit the number of initial recommended words; it can be set to a default value or adjusted as needed. For example, the top-K candidate recommended words can be selected as the initial recommended words, where K=5.

[0038] In this embodiment, during the prediction and analysis process using a preset input recommendation model, semantic discrepancies can be avoided by combining contextual information, and intent tags can achieve accurate matching of recommended words. At the same time, the applicability of the initial recommended words is improved by filtering based on the candidate recommendation probability.

[0039] In one embodiment, step S10, namely obtaining the context information corresponding to the newly added input word, includes: S104. Obtain the session identifier of the newly added input word; S105. Obtain at least one historical conversation text within a preset historical duration based on the conversation identifier, and determine all the historical conversation texts as context information corresponding to the newly added input word.

[0040] Understandably, in text input scenarios of conversational question-and-answer interactions, a unique session identifier is assigned to each session to ensure contextual continuity. The session identifier is the code used to identify a specific interaction session between the server and client, enabling the server to recognize the identities of both parties, track session input states, and maintain a continuous interaction process. Different session identifiers correspond to different identities of the parties involved in the session; for example, different session identifiers may correspond to different dialog boxes. A state machine is maintained to track the state of the dialog box, such as waiting for user input or waiting for a response.

[0041] In one specific embodiment, the server obtains the session identifier of the newly added input word, and determines the dialog box containing the new input word based on the session identifier. It then retrieves at least one historical session text within a preset historical duration from this dialog box, and identifies all historical session texts as context information corresponding to the new input word. Historical session text refers to text already entered in the session before the new input word appears. The preset historical duration is a pre-set maximum duration for extracting text already entered in the session. The preset historical duration can have a default value or be adjusted as needed. For example, if the default value for the preset historical duration is 5 minutes, then the context information corresponding to the new input word refers to text already entered in the session within 5 minutes before the new input word appears in the input box. Furthermore, to avoid excessive context information, a maximum text length can be preset. When the length of the historical session text exceeds the maximum text length, the latest historical session text that meets the maximum text length is retained, and the excess historical session text is removed.

[0042] This embodiment takes into account both session identifiers and preset historical durations to obtain contextual information corresponding to newly added input words. This can avoid mutual interference between different sessions in complex task scenarios and ensure the availability of contextual information.

[0043] In one embodiment, step S30, namely, filtering and sorting all the initial recommended words to obtain a recommended word sequence, includes: S301. Combine each of the initial recommended words with the newly added input words to generate a word group to be verified; S302. Perform verification processing on each of the word groups to be verified according to the preset verification rules, and determine the initial recommended word corresponding to the word group to be verified as the filter recommended word when the verification is passed. S303. Calculate the association similarity between each of the filtered recommended words and the newly added input words using the cosine similarity method, and sort all the filtered recommended words according to the association similarity to obtain a recommended word sequence.

[0044] Understandably, during the filtering and sorting process, the server first combines each initial recommended word with a new input word to generate a word group to be verified. A word group to be verified refers to a group of words formed by combining a new input word first and an initial recommended word last. Each initial recommended word corresponds to one word group to be verified. Next, the server verifies each word group to be verified according to preset verification rules, and if the verification passes, the initial recommended word corresponding to that word group is designated as a filtered recommended word. The preset verification rules are pre-defined standards used to judge whether the combination of two words in a word group meets grammatical requirements, such as grammatical errors, unreasonable part-of-speech collocation, and prohibited words. Filtered recommended words are the initial recommended words remaining after removing words that fail verification; they are a part of the initial recommended words. Finally, the server calculates the association similarity between each filtered recommended word and the new input word using the cosine similarity method, and sorts all filtered recommended words according to the association similarity to obtain a recommended word sequence. Association similarity is a value used to measure the reasonableness of the semantic logical relationship between words. For example, the semantic and logical relationship between "buy" and "insurance" is reasonable, and their correlation similarity is relatively high; while the semantic and logical relationship between "buy" and "disease" is unreasonable, and their correlation similarity is relatively low.

[0045] This embodiment filters out grammatically incorrect or prohibited words through validation, avoiding interference with user selection and improving the quality of text input recommendations. Simultaneously, by calculating the correlation between recommended words and the latest input words using cosine similarity, it prioritizes displaying the most semantically relevant results, thus improving recommendation accuracy.

[0046] In one embodiment, step S40, after determining the target input word from the recommended word sequence according to the recommended word selection instruction, further includes: S401. Generate historical input data corresponding to the target input word based on the newly added input word, context information, and target input word; S402. When input completion is detected, obtain user feedback information and recommendation evaluation results of the input completion text, wherein the input completion text includes multiple target input words; S403. When it is confirmed that the user feedback information and the recommendation evaluation results meet the preset update conditions, the model parameters of the preset input recommendation model are updated according to the historical input data corresponding to all the target input words to obtain the updated preset input recommendation model.

[0047] Understandably, once the server identifies the target input word, it signifies the completion of a new input operation. Based on the new input word, context information, and the target input word, it generates historical input data corresponding to that target input word. Historical input data refers to the actual input word in the completed input operation, along with the previous input word associated with that actual input word and its context information. Each completed new input operation generates a new target input word until the input is complete. Each target input word corresponds to its own set of historical input data.

[0048] Upon detecting input completion, the server retrieves user feedback information and recommendation evaluation results for the completed input text. Input completion refers to the end of continuous text input, such as when the user exits the dialog box or closes the input field. Completed input text refers to all the input text in a continuous text input scenario. For example, in a customer service Q&A interaction, all the text in the dialog box from the beginning to the end of the conversation. Completed input text includes multiple target input words. User feedback information is a quantitative result representing the user's satisfaction with the recommended words selected during text input. Recommendation evaluation results are a quantitative result representing the performance of the pre-set input recommendation model in generating recommended words during text input.

[0049] The server determines whether user feedback and recommendation evaluation results meet preset update conditions. If they do not, the preset input recommendation model does not need to be updated. If they do, the model parameters are updated based on historical input data for all target input words, resulting in an updated model. This updated model is then used in subsequent text input recommendation processes to complete the recommendation. Preset update conditions refer to pre-defined criteria used to determine whether the preset input recommendation model needs parameter updates.

[0050] This embodiment combines the context of the input completed text with preceding vocabulary as sample data for model updates, ensuring that the model can capture new contextual patterns. This embodiment establishes a closed-loop process of user feedback, evaluation, and model updates, enabling it to learn users' personalized habits and dynamically adapt to changes in language and user behavior, thereby providing more relevant recommended words.

[0051] In one embodiment, the user feedback information includes the recommended word selection rate and the recommended word satisfaction, and the recommendation evaluation result includes the recommended word accuracy and the recommendation response time; step S402, namely obtaining the user feedback information and recommendation evaluation result of the input completed text, includes: S4021. Determine the recommended word selection rate of the input completed text based on the target input words in the input completed text and the recommended word sequence corresponding to each target input word; S4022. Determine the satisfaction level of the recommended words for the completed input text based on the user rating results sent by the client; S4023. Determine the recommendation word accuracy of the input completed text based on the initial recommended words and recommended word sequence corresponding to each target input word in the input completed text; S4024. Determine the recommendation response time of the input completion text based on the model recommendation time corresponding to each target input word in the input completion text.

[0052] In step S403, namely confirming that the user feedback information and recommendation evaluation results meet the preset update conditions, the following steps are included: S4031. When the recommended word selection rate is less than a preset selection rate threshold and the recommended word satisfaction is less than a preset satisfaction threshold, confirm that the user feedback information meets the preset update conditions. S4032. When the accuracy of the recommended words is less than a preset accuracy threshold and the recommendation response time is greater than a preset response time threshold, the recommendation evaluation result is confirmed to meet the preset update conditions.

[0053] Understandably, user feedback includes recommended word selection rate and recommended word satisfaction. The completed input text includes multiple target input words; some are selected by the user from the recommended word sequence, while others are manually typed by the user (words in the recommended word sequence do not meet the user's needs). Recommended word selection rate refers to the proportion of target input words selected from the recommended word sequence out of all target input words in the completed input text. Upon detecting input completion, the server sends a rating request to the client regarding satisfaction with the recommended words. The user provides a corresponding rating score (e.g., 1-5 points) through the client, which is then sent to the server as the user rating result. Recommended word satisfaction refers to the rating score in the user rating result.

[0054] The recommendation evaluation results include recommended word accuracy and recommendation response time. For each target input word, a model prediction analysis is performed. The recommended word accuracy for a single target input word is the proportion of the remaining initial recommended words in the recommended word sequence corresponding to that target input word to the total number of initial recommended words before filtering and sorting. The recommended word accuracy for the completed input text is the average of the recommended word accuracies for all individual target input words in the completed input text. For each target input word, a text input recommendation method process is performed. The recommendation response time for a single target input word is the time from detecting a new input word to generating a recommended word sequence within a single text input recommendation method process. The recommendation response time for the completed input text is the average of the recommendation response times for all individual target input words in the completed input text.

[0055] The server performs threshold checks on recommended word selection rate, recommended word satisfaction, recommended word accuracy, and recommended response time to confirm whether preset update conditions are met. If the recommended word selection rate is less than a preset selection rate threshold and the recommended word satisfaction is less than a preset satisfaction threshold, the user feedback is confirmed to meet the preset update conditions. If the recommended word accuracy is less than a preset accuracy threshold and the recommended response time is greater than a preset response time threshold, the recommendation evaluation result is confirmed to meet the preset update conditions. The preset selection rate threshold is a pre-set critical value used to determine whether the preset input recommendation model needs to update its parameters. The preset satisfaction threshold is a pre-set critical value used to determine whether the preset input recommendation model needs to update its parameters. The preset accuracy threshold is a pre-set critical value used to determine whether the preset input recommendation model needs to update its parameters. The preset response time threshold is a pre-set critical value used to determine whether the preset input recommendation model needs to update its parameters.

[0056] This embodiment combines the recommendation word selection rate and recommendation word satisfaction to determine whether user feedback information meets the preset update conditions. On the other hand, it combines the recommendation word accuracy and recommendation response time to determine whether the recommendation evaluation results meet the preset update conditions. This can ensure the necessity and effectiveness of updating the preset input recommendation model and avoid the decrease in model stability and fluctuations in user experience caused by frequent updates.

[0057] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0058] In one embodiment, a text input recommendation device is provided, which corresponds one-to-one with the text input recommendation method in the above embodiments. For example... Figure 3As shown, the text input recommendation device includes an information acquisition module 10, a model prediction module 20, a filtering and ranking processing module 30, and a recommendation word selection module 40. Detailed descriptions of each functional module are as follows: The information acquisition module 10 is used to acquire context information corresponding to the newly added input word when a new input word is detected, and to determine the target intent tag based on the context information. The model prediction module 20 is used to predict and analyze the newly added input words, context information and target intent tags through a preset input recommendation model to obtain at least one initial recommendation word; The filtering and sorting processing module 30 is used to filter and sort all the initial recommended words to obtain a recommended word sequence, and send the recommended word sequence to the client; The recommendation word selection module 40 is used to determine the target input word from the recommendation word sequence according to the recommendation word selection instruction sent by the client when it receives the recommendation word selection instruction corresponding to the recommendation word sequence.

[0059] In one embodiment, the information acquisition module 10 includes: The word segmentation processing unit is used to perform word segmentation processing on the context information to obtain multiple context words; The entity tag lookup unit is used to find the segmented entity tags of each context segmentation from a preset entity tag dictionary; The target intent tag determination unit is used to predict and analyze all the segmented entity tags through a preset intent prediction model to obtain the target intent tag.

[0060] In one embodiment, the model prediction module 20 includes: The candidate recommendation word determination unit is used to predict and analyze the newly added input word, context information and target intent label through a preset input recommendation model to obtain at least one candidate recommendation word and a candidate recommendation probability value corresponding to each candidate recommendation word. The initial recommendation word determination unit is used to select a preset number of initial recommendation words from all the candidate recommendation words based on the candidate recommendation probability value.

[0061] In one embodiment, the information acquisition module 10 further includes: A session identifier determination unit is used to obtain the session identifier of the newly added input word; The context information determination unit is used to obtain at least one historical conversation text within a preset historical time period based on the conversation identifier, and determine all the historical conversation texts as context information corresponding to the newly added input word.

[0062] In one embodiment, the filtering and sorting processing module 30 includes: The word group to be verified generation unit is used to generate word groups to be verified by combining each of the initial recommended words with the newly added input words; The verification processing unit is used to perform verification processing on each of the word groups to be verified according to the preset verification rules, and when the verification is passed, the initial recommended word corresponding to the word group to be verified is determined as the filter recommended word. The recommended word sequence determination unit is used to calculate the association similarity between each of the filtered recommended words and the newly added input words using the cosine similarity method, and to sort all the filtered recommended words according to the association similarity to obtain a recommended word sequence.

[0063] In one embodiment, the recommendation word selection module 40 includes: The historical input data addition unit is used to generate historical input data corresponding to the target input word based on the newly added input word, context information, and target input word. The feedback evaluation information acquisition unit is used to acquire user feedback information and recommendation evaluation results of the input completion text when the input completion is detected, wherein the input completion text includes multiple target input words; The model parameter update unit is used to update the model parameters of the preset input recommendation model based on the historical input data corresponding to all the target input words when it is confirmed that the user feedback information and the recommendation evaluation results meet the preset update conditions, so as to obtain the updated preset input recommendation model.

[0064] In one embodiment, the recommendation word selection module 40 includes a feedback evaluation information acquisition unit comprising: The selection rate determination subunit is used to determine the recommendation word selection rate of the input completed text based on the target input words in the input completed text and the recommendation word sequence corresponding to each target input word. The satisfaction determination subunit is used to determine the satisfaction level of the recommended words in the input completed text based on the user rating results sent by the client. The accuracy determination subunit is used to determine the recommended word accuracy of the input completed text based on the initial recommended words and recommended word sequence corresponding to each target input word in the input completed text; The response time determination subunit is used to determine the recommended response time of the input completed text based on the model recommendation time corresponding to each target input word in the input completed text; The model parameter update unit includes: The first condition judgment subunit is used to confirm that the user feedback information meets the preset update conditions when the recommended word selection rate is less than a preset selection rate threshold and the recommended word satisfaction is less than a preset satisfaction threshold. The second condition judgment subunit is used to confirm that the recommendation evaluation result meets the preset update conditions when the accuracy of the recommended words is less than a preset accuracy threshold and the recommendation response time is greater than a preset response time threshold.

[0065] Specific limitations regarding the text input recommendation device can be found in the limitations of the text input recommendation method above, and will not be repeated here. Each module in the aforementioned text input recommendation 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.

[0066] In one embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 4 As shown, the computer device includes a processor, memory, network interface, and database connected via a system bus. The processor provides computational and control capabilities. The memory includes a readable storage medium and internal memory. The readable storage medium stores an operating system, computer-readable instructions, and a database. The internal memory provides an environment for the operation of the operating system and computer-readable instructions in the readable storage medium. The database stores data related to the text input recommendation method. The network interface communicates with external terminals via a network connection. When the computer-readable instructions are executed by the processor, they implement a text input recommendation method. The readable storage medium provided in this embodiment includes both non-volatile and volatile readable storage media.

[0067] In one embodiment, a computer device is provided, including a memory, a processor, and computer-readable instructions stored in the memory and executable on the processor, wherein the processor performs the following steps when executing the computer-readable instructions: When a new input word is detected, the context information corresponding to the new input word is obtained, and the target intent tag is determined based on the context information; The newly added input words, context information, and target intent tags are predicted and analyzed by a preset input recommendation model to obtain at least one initial recommendation word; All the initial recommended words are filtered and sorted to obtain a recommended word sequence, and the recommended word sequence is sent to the client; Upon receiving a recommendation word selection instruction sent by the client corresponding to the recommendation word sequence, the target input word is determined from the recommendation word sequence according to the recommendation word selection instruction.

[0068] In one embodiment, one or more computer-readable storage media storing computer-readable instructions are provided. The readable storage media provided in this embodiment include non-volatile readable storage media and volatile readable storage media. The readable storage media stores computer-readable instructions, which, when executed by one or more processors, perform the following steps: When a new input word is detected, the context information corresponding to the new input word is obtained, and the target intent tag is determined based on the context information; The newly added input words, context information, and target intent tags are predicted and analyzed by a preset input recommendation model to obtain at least one initial recommendation word; All the initial recommended words are filtered and sorted to obtain a recommended word sequence, and the recommended word sequence is sent to the client; Upon receiving a recommendation word selection instruction sent by the client corresponding to the recommendation word sequence, the target input word is determined from the recommendation word sequence according to the recommendation word selection instruction.

[0069] 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 instructing related hardware with computer-readable instructions. These computer-readable instructions can be stored in a non-volatile readable storage medium or a volatile readable storage medium. When executed, these computer-readable instructions can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided by this invention can include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0070] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is used as an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above.

[0071] The software tools or components not belonging to this company that appear in the embodiments of this application are merely illustrative examples and do not represent actual use. The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. A text input recommendation method, characterized in that, include: When a new input word is detected, the context information corresponding to the new input word is obtained, and the target intent tag is determined based on the context information; The newly added input words, context information, and target intent tags are predicted and analyzed by a preset input recommendation model to obtain at least one initial recommendation word; All the initial recommended words are filtered and sorted to obtain a recommended word sequence, and the recommended word sequence is sent to the client; Upon receiving a recommendation word selection instruction sent by the client corresponding to the recommendation word sequence, the target input word is determined from the recommendation word sequence according to the recommendation word selection instruction.

2. The text input recommendation method as described in claim 1, characterized in that, Determining the target intent tag based on the context information includes: The context information is segmented to obtain multiple contextual words; Find the segmentation entity tags for each context segmentation from the preset entity tag dictionary; The target intent tag is obtained by predicting and analyzing all the segmented entity tags using a preset intent prediction model.

3. The text input recommendation method as described in claim 1, characterized in that, The step involves using a preset input recommendation model to predict and analyze the newly added input words, contextual information, and target intent tags to obtain at least one initial recommended word, including: By using a preset input recommendation model, the newly added input words, context information and target intent tags are predicted and analyzed to obtain at least one candidate recommendation word and a candidate recommendation probability value corresponding to each candidate recommendation word. Based on the candidate recommendation probability value, a preset number of initial recommendation words are selected from all the candidate recommendation words.

4. The text input recommendation method as described in claim 1, characterized in that, The step of obtaining the context information corresponding to the newly added input word includes: Obtain the session identifier of the newly added input word; Based on the session identifier, at least one historical session text within a preset historical duration is obtained, and all the historical session texts are determined as context information corresponding to the newly added input word.

5. The text input recommendation method as described in claim 1, characterized in that, The step of filtering and sorting all the initial recommended words to obtain a recommended word sequence includes: Each initial recommended word is combined with the newly added input word to generate a word group to be verified; The word groups to be verified are verified according to the preset verification rules, and the initial recommended word corresponding to the word group to be verified is determined as the filter recommended word when the verification is passed. The association similarity between each of the filtered recommended words and the newly added input words is calculated using the cosine similarity method, and all the filtered recommended words are sorted according to the association similarity to obtain a recommended word sequence.

6. The text input recommendation method as described in claim 1, characterized in that, After determining the target input word from the recommended word sequence according to the recommended word selection instruction, the method further includes: Based on the newly added input word, context information, and target input word, generate historical input data corresponding to the target input word; When input completion is detected, user feedback information and recommendation evaluation results of the input completion text are obtained, wherein the input completion text includes multiple target input words; When the user feedback information and recommendation evaluation results are confirmed to meet the preset update conditions, the model parameters of the preset input recommendation model are updated according to the historical input data corresponding to all the target input words to obtain the updated preset input recommendation model.

7. The text input recommendation method as described in claim 6, characterized in that, The user feedback information includes the recommended word selection rate and the recommended word satisfaction rate; the recommendation evaluation results include the recommended word accuracy rate and the recommendation response time. The process of obtaining user feedback information and recommendation evaluation results after the input of the completed text includes: Based on the target input words in the input completed text and the recommended word sequence corresponding to each target input word, the recommended word selection rate of the input completed text is determined; The satisfaction level of the recommended words for the completed input text is determined based on the user rating results sent by the client. The recommendation word accuracy of the input completed text is determined based on the initial recommended words and the recommended word sequence corresponding to each target input word in the input completed text. The recommendation response time of the input completion text is determined based on the model recommendation time corresponding to each target input word in the input completion text. The confirmation that the user feedback information and recommendation evaluation results meet the preset update conditions includes: When the selection rate of the recommended words is less than a preset selection rate threshold and the satisfaction level of the recommended words is less than a preset satisfaction level threshold, the user feedback information is confirmed to meet the preset update conditions. When the accuracy of the recommended words is less than a preset accuracy threshold and the recommendation response time is greater than a preset response time threshold, the recommendation evaluation result is confirmed to meet the preset update conditions.

8. A text input recommendation device, characterized in that, include: The information acquisition module is used to acquire context information corresponding to the newly added input word when a new input word is detected, and to determine the target intent tag based on the context information; The model prediction module is used to predict and analyze the newly added input words, context information and target intent tags through a preset input recommendation model to obtain at least one initial recommendation word. The filtering and sorting processing module is used to filter and sort all the initial recommended words to obtain a recommended word sequence, and send the recommended word sequence to the client; The recommended word selection module is used to determine the target input word from the recommended word sequence according to the recommended word selection instruction sent by the client when it receives the recommended word selection instruction corresponding to the recommended word sequence.

9. A computer device comprising a memory, a processor, and computer-readable instructions stored in the memory and executable on the processor, characterized in that, When the processor executes the computer-readable instructions, it implements the text input recommendation method as described in any one of claims 1 to 7.

10. A computer-readable storage medium storing computer-readable instructions, characterized in that, When the computer-readable instructions are executed by one or more processors, the one or more processors cause the one or more processors to perform the text input recommendation method as described in any one of claims 1 to 7.

Images