Intelligent question-answering method for exhibition and exhibition robot

Abstract

The application relates to the technical field of computer application, and provides an intelligent question and answer method for a recruitment exhibition and an exhibition robot. The method obtains user question data and corresponding standard answer data under a historical recruitment exhibition scene, and obtains participant background information and exhibition basic information associated with the historical question information; the obtained user question data, the standard answer data, the participant background information and the exhibition basic information are input into an intelligent question and answer model for training; in a recruitment exhibition site, real-time question information of a user is obtained through an input module of the exhibition robot, real-time background information of a current question user and exhibition real-time dynamic information are collected, the obtained real-time question information, the real-time background information and the exhibition real-time dynamic information are input into the trained intelligent question and answer model, and personalized answer content for the current user is generated; and the generated personalized answer content is fed back to the user in the form of voice or text through an output module of the exhibition robot.

Description

Technical Field

[0001] This application relates to the field of computer application technology, and in particular to an intelligent question-answering method and exhibition robot for trade fairs. Background Technology

[0002] In trade fair scenarios, the intelligent question-and-answer function of exhibition robots is an important technological means to enhance the attendee experience and improve exhibition efficiency. Traditional exhibition question-and-answer methods are usually based on fixed knowledge bases or keyword matching technology, which can only provide standardized answers and cannot generate targeted answers based on the attendee's personalized background (such as identity type, job intention) and real-time exhibition dynamics (such as changes in booth activities and adjustments to job requirements). Existing technologies have the following shortcomings:

[0003] 1. Limited data dimensions: The model relies solely on pre-set basic exhibition information (such as booth distribution and event schedule) without combining historical user question data and attendee background information for model training, thus failing to learn the characteristics of user needs in real-world scenarios.

[0004] 2. Poor dynamic adaptability: Unable to collect temporary change information at the exhibition site in real time (such as sudden adjustments to job requirements or changes in event time), and the response content lags behind the actual situation of the exhibition;

[0005] 3. Lack of personalization: Ignoring the real-time background information of participants (such as education, work experience, and job intentions), the answers lack specificity and fail to meet the differentiated needs of different types of users (such as students and job seekers).

[0006] Currently, no solution combines historical question data, attendee background information, basic exhibition information, and real-time dynamic data to achieve personalized question and answer through intelligent model training.

[0007] Therefore, a method is urgently needed to solve at least one of the above problems. Summary of the Invention

[0008] This application provides an intelligent question-answering method and exhibition robot for trade fairs, aiming to solve the problem that there is no existing solution that combines historical question data, participant background information, basic exhibition information and real-time dynamic data to achieve personalized question-answering through intelligent model training.

[0009] Firstly, this application provides an intelligent question-answering method for trade fairs, applied to trade fair robots, including:

[0010] The system acquires user question data and corresponding standard answer data from historical job fairs, and obtains participant background information and basic exhibition information associated with the historical question information. The user question data includes historical questions from different types of participants regarding the exhibition process, booth information, job requirements, and event arrangements.

[0011] The acquired user question data, standard answer data, participant background information, and basic exhibition information are input into the intelligent question answering model for training. By adjusting the parameters of the intelligent question answering model, the corresponding answer content is generated, thus completing the training of the intelligent question answering model.

[0012] At the trade fair, the input module of the exhibition robot obtains real-time questions from users, collects real-time background information of the current questioning user and real-time dynamic information of the exhibition. The real-time background information includes the user's on-site registered identity information and job intentions, and the real-time dynamic information of the exhibition includes updates on booth activities and temporarily changed recruitment job requirements for the current period.

[0013] The acquired real-time question information, real-time background information, and real-time exhibition dynamic information are input into the trained intelligent question-answering model to generate personalized answers for the current user; the generated personalized answers are then fed back to the user in voice or text form through the output module of the exhibition robot.

[0014] Secondly, this application also provides a convention and exhibition robot, including:

[0015] Memory and processor;

[0016] The memory is used to store computer programs;

[0017] The processor is configured to execute the computer program and, in executing the computer program, implement the steps of the intelligent question-and-answer method for trade fairs as described in the first aspect above.

[0018] This application provides an intelligent question-answering method and exhibition robot for trade fairs. The method integrates historical user question data, standard answer data, participant background information, and basic exhibition information, enabling the intelligent question-answering model to learn user demand patterns in real-world scenarios, thus improving the accuracy and relevance of answers. By collecting participant background information and exhibition dynamic information (such as booth activity updates and job requirement changes) in real time, the method ensures that answer content adapts to changes at the exhibition, avoiding the lag of a fixed knowledge base. Customized answers are generated based on real-time user background (such as identity type and job intention), providing differentiated information for different user groups (such as recent graduates and experienced professionals), significantly improving participant experience and exhibition service efficiency. Through dynamic input of real-time data, the model can implicitly learn the temporary changing patterns of the exhibition scenario, accumulating richer training data for subsequent similar scenarios and forming a technological iteration advantage.

[0019] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and do not limit this application. Attached Figure Description

[0020] To more clearly illustrate the technical solutions of the embodiments of this application, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0021] Figure 1 This is a schematic flowchart illustrating the steps of an intelligent question-and-answer method for trade fairs provided in one embodiment of this application;

[0022] Figure 2 This is a schematic diagram of the structure of an intelligent question-and-answer system for trade fairs provided in one embodiment of this application;

[0023] Figure 3 This is a schematic block diagram of the structure of a convention robot provided in one embodiment of this application.

[0024] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and do not limit this application. Detailed Implementation

[0025] 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, 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.

[0026] The flowchart shown in the attached diagram is for illustrative purposes only and does not necessarily include all content and operations / steps, nor does it necessarily have to be performed in the order described. For example, some operations / steps can be broken down, combined, or partially merged, so the actual execution order may change depending on the actual situation.

[0027] It should be understood that, in order to clearly describe the technical solutions of the embodiments of the present invention, the terms "first" and "second" are used in the embodiments of the present invention to distinguish identical or similar items with essentially the same function and effect. Those skilled in the art will understand that the terms "first" and "second" do not limit the quantity or execution order, and the terms "first" and "second" are not necessarily different.

[0028] It should be understood that the terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the scope of the application. As used in this specification and the appended claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms unless the context clearly indicates otherwise.

[0029] It should also be understood that the term “and / or” as used in this application specification and the appended claims means any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.

[0030] The following detailed description of some embodiments of this application is provided in conjunction with the accompanying drawings. Unless otherwise specified, the following embodiments and features can be combined with each other.

[0031] In trade fair scenarios, the intelligent question-and-answer function of exhibition robots is an important technological means to enhance the attendee experience and improve exhibition efficiency. Traditional exhibition question-and-answer methods are usually based on fixed knowledge bases or keyword matching technology, which can only provide standardized answers and cannot generate targeted answers based on the attendee's personalized background (such as identity type, job intention) and real-time exhibition dynamics (such as changes in booth activities and adjustments to job requirements). Existing technologies have the following shortcomings:

[0032] The data dimension is limited: it only relies on preset basic exhibition information (such as booth distribution and event schedule), without combining historical user question data and attendee background information for model training, and is unable to learn the user demand characteristics in real scenarios;

[0033] Poor dynamic adaptability: Unable to collect temporary change information at the exhibition site in real time (such as sudden adjustments to job requirements or changes in event time), and the response content lags behind the actual situation of the exhibition.

[0034] Lack of personalization: Ignoring the real-time background information of participants (such as education, work experience, and job intentions), the answers lack specificity and fail to meet the differentiated needs of different types of users (such as students and job seekers).

[0035] Currently, no existing technology combines historical question data, participant background information, basic exhibition information, and real-time dynamic data to achieve personalized question-and-answer through intelligent model training. Therefore, how to train an intelligent question-and-answer model using multi-dimensional data and generate dynamic, personalized answers based on real-time scenario information is a pressing technical problem to be solved in the field of intelligent question-and-answer for trade fairs.

[0036] To resolve the above issues, please refer to [link / reference]. Figure 1 , Figure 1 This is a schematic flowchart illustrating an intelligent question-and-answer method for trade fairs provided in one embodiment of this application. This intelligent question-and-answer method for trade fairs can be implemented by a trade fair robot.

[0037] It should be noted that the acquisition of any information involved in the provided methods is in compliance with relevant regulations and is carried out with the user's consent, and will not infringe on the user's privacy or violate relevant laws and regulations.

[0038] Specifically, such as Figure 1 As shown, the intelligent question-answering method provided for trade fairs includes steps S101 to S104, which are detailed below:

[0039] Step S101. Obtain user question data and corresponding standard answer data from historical job fair scenarios, and obtain participant background information and basic exhibition information associated with historical question information; the user question data includes historical question information from different types of participants regarding the exhibition process, booth information, job requirements, and event arrangements.

[0040] Specifically, the core of this step is to collect and structure multi-dimensional data from historical trade fair scenarios to provide rich input features for model training. This includes four types of data: User Question Data: Covering different types of historical questions, such as trade fair procedures (e.g., check-in process, document processing), booth information (e.g., booth location, list of exhibitors), job requirements (e.g., education, major, work experience), and event arrangements (e.g., presentation times, interactive sessions), in text and speech-to-text formats. Standard Answer Data: Standardized answers to historical questions, which may include text, links, attachments (e.g., job descriptions), and must correspond one-to-one with the question data. Attendees Background Information: Structured data related to the identity characteristics of users who asked questions in the past, such as education (Bachelor's / Master's / Doctoral), work experience (recent graduate / 3+ years experience / no experience), job intentions (job type, salary expectations, industry preference), and identity type (student / employed job seeker / freelancer). Basic exhibition information: Static exhibition information, such as booth distribution map, event schedule, list of exhibitors, list of job requirements (including fixed job descriptions) and other structured data.

[0041] Data is extracted from historical exhibition log systems (such as user interaction records and customer service dialogue logs), participant registration forms (including background information fields), and exhibition management systems (which obtain basic data on booths, activities, and positions).

[0042] Data cleaning and structuring utilizes Natural Language Processing (NLP) techniques to perform entity recognition (e.g., extracting "booth number" and "job title") and intent classification on unstructured question text; standardized encoding of participant background information (e.g., converting education level into numerical labels); and linking question data with background information using unique identifiers (e.g., session IDs). Data acquisition requires user consent, sensitive information (e.g., contact information) is anonymized, and encryption is used during storage to ensure compliance with relevant regulations.

[0043] Step S102. Input the acquired user question data, standard answer data, participant background information and exhibition basic information into the intelligent question answering model for training. Generate corresponding answer content by adjusting the parameters of the intelligent question answering model to complete the training of the intelligent question answering model.

[0044] Specifically, the model is trained based on multi-dimensional historical data, enabling it to learn the mapping relationship between user needs, background information, and answers. The model needs to integrate structured (background information, basic exhibition data) and unstructured (question text, answer text) data, and support dynamic input of real-time scene information.

[0045] The model architecture employs a hybrid approach, such as an encoder-decoder framework (e.g., Transformer) combined with a structured data embedding layer: The text encoder semantically encodes the user's question text (e.g., using a BERT pre-trained model) to extract contextual semantic features. The structured data processing layer converts attendee background information (e.g., education level, job aspirations) and basic exhibition information (e.g., booth ID, job type) into numerical vectors (e.g., one-hot encoding, embedding vectors), which are then fused with textual features through a fully connected layer. An attention mechanism is introduced to enable the model to focus on key entities in the question (e.g., "R&D position," "experience requirements") and generate targeted answers based on background information.

[0046] The training data was constructed by concatenating the question text, background information vector, and basic exhibition information vector as input features, and using the standard answer as the output label. Data augmentation techniques (such as synonym replacement and question sentence transformation) were employed to expand the sample size and improve the model's generalization ability.

[0047] The training process employs supervised learning methods, using cross-entropy (for classification tasks) or sequence generation loss (for text generation tasks) as the loss function. The optimization objective is to minimize the difference between the predicted answer and the standard answer. Training is conducted in stages: first, pre-training is performed on a general question-answering dataset, then fine-tuning is done based on data from the trade fair domain to improve domain adaptability. Model evaluation is conducted through accuracy, BLEU score (for text generation tasks), and user satisfaction simulation tests to verify the model's ability to handle multi-dimensional data, iteratively adjusting parameters (such as learning rate and attention weights).

[0048] Step S103. At the exhibition site, the real-time question information of users is obtained through the input module of the exhibition robot. The real-time background information of the current questioning user and the real-time dynamic information of the exhibition are collected. The real-time background information includes the identity information and job intention registered by the user on site. The real-time dynamic information of the exhibition includes the current booth activity updates and temporarily changed recruitment job requirements.

[0049] Specifically, user interaction data and dynamic scene information are collected in real time at the exhibition to provide real-time input for personalized answers. This includes: Real-time question information: The current question input by the user via voice or text needs to be converted into structured text (e.g., speech-to-text, removal of invalid characters). Real-time background information: The identity information registered by the user on-site (e.g., education level, years of work experience obtained by scanning a QR code and filling out a form) and job intentions (e.g., desired position, industry) form a real-time user profile. Real-time dynamic information of the exhibition: Real-time data such as temporary changes to booth activities (e.g., postponement of a booth's presentation) and adjustments to job requirements (e.g., addition of urgent recruitment positions, modification of salary ranges) needs to be synchronized with the exhibition management system in real time.

[0050] The input module design includes: Interaction Interface: User questions are received via the exhibition robot's microphone (speech recognition), touchscreen (text input), or mobile app. ASR (Automatic Speech Recognition) technology is used to convert the questions into text, and NLP technology is combined to analyze the intent of the question (e.g., inquiring about job positions, querying event times). Background Information Collection: Upon the user's first interaction, real-time background information is obtained through a short form (e.g., "What is your educational background?", "What position do you expect to hold?") or by scanning a QR code. This information is stored as structured fields (e.g., JSON format).

[0051] Dynamic data synchronization is achieved by establishing a real-time API interface with the exhibition organizer's management system (such as ERP or booth management platform) to periodically retrieve data (e.g., once per minute) or subscribe to change events (e.g., push notifications when job requirements are updated) to obtain the real-time status of booth activities and job requirements. Dynamic information is validated (e.g., for valid time formats and complete job requirements) to ensure the accuracy of the data input into the model.

[0052] Step S104. Input the acquired real-time question information, real-time background information, and real-time exhibition dynamic information into the trained intelligent question-answering model to generate personalized answer content for the current user; and feed back the generated personalized answer content to the user in voice or text form through the output module of the exhibition robot.

[0053] Based on real-time input data and a trained model, personalized answers tailored to the user's background and the dynamics of the exhibition are generated and fed back through multimodal methods. Core technologies include: Multi-dimensional input fusion: Inputting real-time question text, user background vectors, and exhibition dynamic information (such as changed event times and new job descriptions) into the model triggers its dynamic response mechanism. Answer generation strategy: Adjusting the focus of the answer based on the user's identity (e.g., students focusing on internship processes, while employed individuals focus on salary and promotion), and incorporating real-time dynamic information (e.g., prioritizing the latest times when events change). Multimodal output: Converting text answers into speech (TTS technology) or text display, supporting additional information such as images and links (e.g., job QR codes, event maps).

[0054] Input preprocessing involves segmenting the real-time question text into words and recognizing entities (e.g., extracting "booth 3" and "Java development position"), and concatenating them with background information (e.g., "Master's degree" and "3 years of development experience") and dynamic information (e.g., "booth 3 has added a Java position, requiring 3 years of experience") to form the model input vector.

[0055] The structured data in the exhibition's dynamic information (such as "salary range" and "experience requirements" after the adjustment of job requirements) is normalized and converted into a format consistent with the training data.

[0056] Model inference and response generation utilize a pre-trained model for forward propagation, outputting response text. For text generation models (such as Seq2Seq), an optimal sequence is generated using a beam search algorithm; for retrieval models, the best response is matched from a candidate answer database based on real-time data, and the content is dynamically adjusted (e.g., inserting real-time event times). A rule engine is introduced to validate the responses, ensuring they contain the latest dynamic information (e.g., prioritizing real-time job requirements over historical, fixed data).

[0057] Multimodal feedback includes: Voice output: Text-to-speech (TTS) technology converts text responses into natural speech, supporting dialect recognition and synthesis (adjusted according to the user's geographic background). Text display: Responses are displayed on the robot's screen or the user's mobile device, with key information (such as booth number and time) highlighted and clickable links (such as redirection to job details). Feedback format is dynamically adjusted based on user interaction history (e.g., voice responses prioritize high-frequency questions, while text + charts are prioritized for complex information). Responses do not contain unauthorized private information (such as contact information not provided during registration). Historical interaction data is only used for the current session and is deleted or anonymized after the session ends according to rules.

[0058] In some embodiments, the standard answer data includes standard answer content for various types of questions, the participant background information includes the participant's identity type and job search direction, and the exhibition basic information includes the exhibition booth distribution, the list of recruiting companies, and the event schedule. The process of obtaining user question data and corresponding standard answer data from historical job fair scenarios, and obtaining participant background information and exhibition basic information associated with historical question information, includes: extracting user question data and corresponding standard answer data from the user interaction log database of historical job fairs through a data interface; obtaining the corresponding identity type and job search direction from the participant registration database based on the user ID; extracting booth distribution coordinate data, the list of recruiting companies, and the event schedule associated with the booth number and event name involved in the historical questions from the exhibition management system database; and classifying and storing the data according to the question type.

[0059] This embodiment clarifies the specific method for obtaining multi-dimensional historical data in step S101. It extracts and associates user questions, background information and exhibition basic information from three types of databases through data interfaces to achieve data structuring and classified storage.

[0060] Data sources and extraction include: User questions and standard answers: extracted from the user interaction log database of historical job fairs via a data interface. The logs contain user IDs, question texts, timestamps, and corresponding standard answers (such as text and links). Participant background information is obtained from the participant registration database based on user IDs, including structured fields such as identity type (student / employed / other) and job search direction (job type, industry preference). Basic exhibition information is extracted from the exhibition management system database, using entities such as "booth number" and "event name" mentioned in historical questions to obtain booth distribution coordinates, a list of recruiting companies, and an event schedule (such as the company name corresponding to the booth number and the event date / time).

[0061] Data processing and storage involves categorizing extracted data by question type (exhibition process, booth information, etc.) and establishing a tagging system (e.g., the tag "booth location" corresponds to records where the question text contains "booth number"). Data is stored using relational databases (such as MySQL) or document-oriented databases (such as MongoDB), with primary keys such as user ID and question ID ensuring data consistency.

[0062] In some embodiments, the process of inputting the acquired user question data, standard answer data, participant background information, and exhibition basic information into the intelligent question answering model for training, and generating corresponding answer content by adjusting the parameters of the intelligent question answering model to complete the training of the intelligent question answering model, includes: performing natural language processing on the user question data to extract keywords and semantic vectors; converting the participant background information and exhibition basic information into structured feature vectors; concatenating the above vectors as the input to the intelligent question answering model; using a deep learning-based sequence-to-sequence model or a retrieval-generation hybrid model as the intelligent question answering model, with the standard answer data as the target output; optimizing the model parameters through a backpropagation algorithm to ensure that the semantic similarity between the answer content output by the intelligent question answering model and the standard answer reaches a preset threshold; during the training process, adding an attention mechanism to strengthen the association weight between the question keywords and the exhibition basic information and participant background information; dividing the training set and validation set through a cross-validation method; and iteratively adjusting the hyperparameters of the intelligent question answering model until the model converges.

[0063] This embodiment details the model training process in step S102, including natural language processing, structured feature transformation, model architecture selection (sequence-to-sequence or retrieval-generation hybrid model), and training optimization methods (attention mechanism, cross-validation).

[0064] Input data processing includes: Text processing: segmenting user queries, tagging parts of speech, and extracting keywords (such as TF-IDF), generating semantic vectors using pre-trained models such as BERT. Structured feature transformation: converting identity type (one-hot encoding), job search direction (embedding vector), booth coordinates (numerical features), etc., into feature vectors of a unified dimension, which are then concatenated with the semantic vectors as model input.

[0065] Model architecture and training include: Model selection: Sequence-to-sequence model (Seq2Seq): suitable for open-domain generation, the encoder processes the input vector, and the decoder generates the response text. Retrieval-generation hybrid model: First, the retrieval module matches candidate answers from the answer database, and then the generation module fine-tunes the content to improve efficiency.

[0066] The optimization mechanism introduces an attention mechanism, enabling the model to dynamically weight question keywords and exhibition information when generating answers (e.g., for questions about "job requirements," priority is given to job descriptions from the list of recruiting companies). The training and validation sets are divided in an 8:2 ratio. Cross-validation is used to evaluate the model's accuracy on different question categories. Hyperparameters such as the learning rate and hidden layer dimensions are iteratively adjusted until the validation set loss converges.

[0067] In some embodiments, obtaining real-time user questions through the input module of the exhibition robot includes: collecting user voice signals through the microphone of the input module, converting the voice into text data using speech recognition technology; performing word segmentation, stop word removal, and grammatical error correction on the converted text data, identifying the question type through an intent classification model, and generating structured real-time question information; the question types include at least exhibition process, booth information, job requirements, and event arrangements.

[0068] This embodiment clarifies the process of obtaining real-time question information in step S103, including speech recognition, text cleaning and intent classification, and generating structured question types (exhibition process type, booth information type, etc.).

[0069] The exhibition robot collects voice signals through a microphone, converts them into text using ASR (Automatic Speech Recognition) technology (such as Baidu Voice and iFlytek API), and supports noise reduction processing to improve recognition accuracy.

[0070] Text preprocessing and intent classification include: Cleaning steps: Word segmentation (such as jieba word segmentation), removing stop words ("of", "呢"), and grammar correction (based on rules or pre-trained correction models). Intent classification: Using a pre-trained classification model (such as FastText) or a custom rule engine, identify the question type according to keywords ("booths" corresponding to "booth information category", "process" corresponding to "exhibition process category"), and output a structured label (such as {"type": "job requirement category", "content": "How many years of experience are required for a Java development position?"}).

[0071] In some embodiments, collecting the real-time background information of the current question-asking user and the real-time dynamic information of the exhibition includes: Obtaining the identity information and job-hunting intention registered by the user on-site through the touch screen or barcode scanning device carried by the exhibition robot, and generating real-time background information data; Obtaining the booth activity update data and the temporarily changed recruitment position requirement data of the current period from the exhibition real-time management system through a network interface, and structuring and storing the real-time dynamic information of the exhibition according to the timestamp and booth number.

[0072] This embodiment details the acquisition method of real-time background and dynamic information in step S103, obtains the user background through a touch screen / barcode scanning device, synchronizes the booth activities and position requirement changes in real time from the exhibition management system, and stores them structurally according to the timestamp and booth number.

[0073] Real-time background information collection includes: User interaction: The user fills in a form (such as "identity type", "expected position") through the robot touch screen or scans a code to jump to a mini-program for registration, and the data is synchronously stored locally in the robot in real time, generating JSON format data containing identity information (education background, working years) and job-hunting intention (salary expectation, industry).

[0074] Obtaining the exhibition dynamic information is connected to the exhibition real-time management system through a RESTful API, pulling data regularly (such as every 5 minutes) or subscribing to Webhook to receive change notifications. The data obtained includes: Booth activity updates (such as "The lecture time at Booth 2 is adjusted from 10:00 to 10:30", with the timestamp "2025-08-29T11:00:00"); Position requirement changes (such as "A new 'front-end development' position is added at Booth 5, requiring 3 years of experience", with the booth number and change time). Establish an index according to "timestamp + booth number" and store it in an in-memory database (such as Redis) or a real-time data warehouse to ensure low-latency access.

[0075] In some embodiments, the step of inputting the acquired real-time question information, real-time background information, and real-time exhibition dynamic information into a trained intelligent question-answering model to generate personalized answer content for the current user includes: concatenating the semantic vector of the real-time question information, the structured feature vector of the real-time background information, and the timestamp and change feature vector of the real-time exhibition dynamic information as input to the trained intelligent question-answering model; the intelligent question-answering model prioritizes matching recruitment job requirements and booth activity information related to the user's job search intentions based on an attention mechanism to generate initial answer content; post-processing the initial answer content, adjusting the expression according to the user's identity type, embedding temporary changes in the real-time exhibition dynamic information, and forming final answer content that includes the user's personalized needs.

[0076] This embodiment clarifies the generation logic of personalized answers in step S104, including multi-dimensional vector concatenation, attention mechanism matching job intentions, post-processing embedding dynamic information, and adjusting the expression according to the user's identity.

[0077] The input vector is constructed by concatenating the semantic vector of the real-time question (generated by BERT), the real-time background feature vector (one-hot encoding of identity type + embedding vector of job direction), and the dynamic information features (time stamp converted to time encoding, binary encoding of change type) into the model input tensor.

[0078] Answer generation and post-processing include: Attention mechanism: When generating answers, the model prioritizes information about exhibitions that are relevant to the user's job search intentions (e.g., when the job search direction is "R&D", it focuses on matching R&D job descriptions in the list of recruiting companies). Dynamic information embedding: If the dynamic information includes changes in booth activity times, the old time is replaced in the answer first (e.g., "The original 10:00 presentation has been moved to 10:30, please go to booth number 2"). For student users, the language style focuses more on "internship process" and "resume submission method", while for employed users, it emphasizes "salary structure" and "career advancement path".

[0079] In some embodiments, the personalized response content generated by the output module of the exhibition robot is fed back to the user in the form of voice or text, including: if voice feedback is selected, the text response content is converted into natural speech through a speech synthesis engine, and the voice parameters are adjusted according to the user's language habits and speech rate preferences; if text feedback is selected, the response content is displayed on the robot's touch screen in a preset format, and interactive booth navigation links or job details links are generated for the user to click and view.

[0080] This embodiment refines the multimodal feedback mechanism of step S104, including the interactive design of speech synthesis parameter adjustment (language habits, speech rate) and text feedback (link generation, format optimization).

[0081] Voice feedback uses a TTS (text-to-speech) engine (such as Google Cloud TTS) to convert the response text into speech, selects the speaker based on the user's registered language habits (such as Mandarin / Cantonese), and supports speech speed adjustment (default 1.0x speed, which users can adjust through interactive settings).

[0082] Text feedback is displayed on the robot's touchscreen in a preset format, with key information (booth number, time, job title) highlighted in a color such as red. Interactive links are generated: clicking "booth navigation" redirects to the exhibition map page (displaying the corresponding booth coordinates), and clicking "job details" loads the job description page from the recruiting company's official website, supporting embedded display on H5 pages.

[0083] In some embodiments, the method further includes: during the trade fair, collecting user feedback data on the answers in real time, and incrementally training the intelligent question-answering model through an online learning algorithm; when the frequency of booth activity updates or job posting changes in the real-time dynamic information corresponding to the trade fair exceeds a preset threshold, triggering an adaptive adjustment mechanism for the model parameters, prioritizing the updating of feature weights related to the changed information, so that the intelligent question-answering model outputs real-time adaptations to the dynamic changes at the trade fair.

[0084] This embodiment adds an online learning and dynamic adaptation mechanism, collects user feedback in real time (such as the "Is this helpful?" button), optimizes the model through incremental training, and triggers adaptive parameter adjustment when dynamic information changes frequently.

[0085] The online learning process involves setting up feedback buttons ("helpful" / "useless") on the robot's interactive interface. After a user clicks, feedback data (including the question text, answer content, and feedback time) is recorded and synchronized to the model training platform periodically (e.g., hourly). Incremental learning algorithms (such as online gradient descent) are used to add new feedback data to the training set, fine-tune model parameters, and focus on optimizing question types that are frequently misclassified.

[0086] The dynamic adaptive mechanism triggers a model parameter reset process when the frequency of changes to booth activities or job requirements in the exhibition management system exceeds a preset threshold (e.g., 10 times / hour): it prioritizes updating the feature weights related to the changed information (e.g., attention weights for fields such as "activity time" and "job requirements"); and temporarily increases the dimensionality of dynamic information in the input vector to ensure that the model output reflects the latest changes first.

[0087] In some embodiments, the method further includes: constructing an answer quality assessment model, calculating assessment indicators such as semantic similarity, information completeness, and timeliness scores based on standard answers and actual answer content in historical user question data; after generating personalized answer content, performing real-time verification of the answer content through the answer quality assessment model; if the assessment indicators are lower than preset standards, triggering a candidate answer generation strategy, selecting the candidate answer with the highest matching degree with the user's real-time background information and the real-time dynamic information of the exhibition from a preset answer template library, and replacing the initially generated answer content.

[0088] This embodiment introduces an answer quality assessment model, which verifies answers using three indicators: semantic similarity, information completeness, and timeliness. If the answer is below the standard, an alternative answer strategy is triggered, and the answer with the highest matching degree is selected from the template library.

[0089] The evaluation metrics include: Semantic similarity: using cosine similarity to calculate the semantic vector distance between the generated answer and the standard answer, with a threshold of ≥0.8; Information completeness: checking whether the answer contains the key entities in the question (e.g., if the question is "the event time of booth 3", the answer must include the specific time), and points will be deducted if they are missing; Timeliness: comparing the timestamps of dynamic information involved in the answer with the current time, and marking it as "outdated" if it has not been updated for more than 30 minutes.

[0090] Alternative response strategies include: a pre-set response template library, categorized by question type, user identity, and dynamic information type (e.g., the "Student - Job Requirements - Temporarily Added Job" template); when evaluation indicators fail to meet the standards, templates are filtered through rule matching (e.g., identity type + question type), populated with real-time dynamic data (e.g., new job name, latest activity time), and the initial response is replaced.

[0091] Please see Figure 2 As shown, Figure 2 This is a schematic diagram of the structure of an intelligent question-and-answer system 200 for trade fairs provided in this application embodiment. The intelligent question-and-answer system 200 for trade fairs is used to execute the steps of the intelligent question-and-answer method for trade fairs shown in the above embodiments. The intelligent question-and-answer system 200 for trade fairs can be a single server or a server cluster, or it can be a terminal, such as a handheld terminal, a laptop computer, a wearable device, or a robot.

[0092] like Figure 2 As shown, the intelligent question-and-answer system 200 used for trade fairs includes:

[0093] The data acquisition unit 201 is used to acquire user question data and corresponding standard answer data from historical job fair scenarios, and to acquire participant background information and basic exhibition information associated with historical question information; the user question data includes historical question information from different types of participants regarding the exhibition process, booth information, job requirements and event arrangements;

[0094] The model training unit 202 is used to input the acquired user question data, standard answer data, participant background information and exhibition basic information into the intelligent question answering model for training. By adjusting the corresponding parameters of the intelligent question answering model, the corresponding answer content is generated to complete the training of the intelligent question answering model.

[0095] The dynamic update unit 203 is used to obtain real-time question information from users through the input module of the exhibition robot at the exhibition site, and to collect real-time background information of the current questioning user and real-time dynamic information of the exhibition. The real-time background information includes the user's on-site registered identity information and job intention, and the real-time dynamic information of the exhibition includes the current booth activity updates and temporarily changed recruitment job requirements.

[0096] The response feedback unit 204 is used to input the acquired real-time question information, real-time background information, and real-time dynamic information of the exhibition into the trained intelligent question-answering model to generate personalized response content for the current user; the generated personalized response content is fed back to the user in voice or text form through the output module of the exhibition robot.

[0097] It should be noted that those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the intelligent question-and-answer system and its modules described above for trade fairs can be referred to the corresponding processes in the embodiments of the intelligent question-and-answer method for trade fairs described above, and will not be repeated here.

[0098] The aforementioned intelligent question-answering method for trade fairs can be implemented as a computer program, which can, for example... Figure 2 It runs on the system shown.

[0099] Please see Figure 3 , Figure 3 This is a schematic block diagram of the structure of an exhibition robot provided in an embodiment of this application. The exhibition robot includes a processor, a memory, and a network interface connected via a device bus, wherein the memory may include a storage medium and internal memory.

[0100] The storage medium may store operating devices and computer programs. The computer program includes program instructions that, when executed, cause the processor to perform any intelligent question-and-answer method for trade shows.

[0101] The processor provides computing and control capabilities to support the operation of the entire exhibition robot.

[0102] Internal memory provides an environment for the execution of computer programs in non-volatile storage media. When the computer program is executed by the processor, it enables the processor to execute any intelligent question-and-answer method used for trade fairs.

[0103] This network interface is used for network communication, such as sending assigned tasks. Those skilled in the art will understand that... Figure 3 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the terminal to which the present application is applied. A specific exhibition robot may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0104] It should be understood that the processor can be a Central Processing Unit (CPU), but it can also be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. Among these, a general-purpose processor can be a microprocessor or any conventional processor.

[0105] In one embodiment, the processor is configured to run a computer program stored in memory to perform the following steps:

[0106] The system acquires user question data and corresponding standard answer data from historical job fairs, and obtains participant background information and basic exhibition information associated with the historical question information. The user question data includes historical questions from different types of participants regarding the exhibition process, booth information, job requirements, and event arrangements.

[0107] The acquired user question data, standard answer data, participant background information, and basic exhibition information are input into the intelligent question answering model for training. By adjusting the parameters of the intelligent question answering model, the corresponding answer content is generated, thus completing the training of the intelligent question answering model.

[0108] At the trade fair, the input module of the exhibition robot obtains real-time questions from users, collects real-time background information of the current questioning user and real-time dynamic information of the exhibition. The real-time background information includes the user's on-site registered identity information and job intentions, and the real-time dynamic information of the exhibition includes updates on booth activities and temporarily changed recruitment job requirements for the current period.

[0109] The acquired real-time question information, real-time background information, and real-time exhibition dynamic information are input into the trained intelligent question-answering model to generate personalized answers for the current user; the generated personalized answers are then fed back to the user in voice or text form through the output module of the exhibition robot.

[0110] In some embodiments, the standard answer data includes standard answer content for various types of questions, the participant background information includes the participant's identity type and job search direction, and the exhibition basic information includes the exhibition booth distribution, the list of recruiting companies, and the event schedule. The process of obtaining user question data and corresponding standard answer data from historical job fair scenarios, and obtaining participant background information and exhibition basic information associated with historical question information, includes: extracting user question data and corresponding standard answer data from the user interaction log database of historical job fairs through a data interface; obtaining the corresponding identity type and job search direction from the participant registration database based on the user ID; extracting booth distribution coordinate data, the list of recruiting companies, and the event schedule associated with the booth number and event name involved in the historical questions from the exhibition management system database; and classifying and storing the data according to the question type.

[0111] In some embodiments, the process of inputting the acquired user question data, standard answer data, participant background information, and exhibition basic information into the intelligent question answering model for training, and generating corresponding answer content by adjusting the parameters of the intelligent question answering model to complete the training of the intelligent question answering model, includes: performing natural language processing on the user question data to extract keywords and semantic vectors; converting the participant background information and exhibition basic information into structured feature vectors; concatenating the above vectors as the input to the intelligent question answering model; using a deep learning-based sequence-to-sequence model or a retrieval-generation hybrid model as the intelligent question answering model, with the standard answer data as the target output; optimizing the model parameters through a backpropagation algorithm to ensure that the semantic similarity between the answer content output by the intelligent question answering model and the standard answer reaches a preset threshold; during the training process, adding an attention mechanism to strengthen the association weight between the question keywords and the exhibition basic information and participant background information; dividing the training set and validation set through a cross-validation method; and iteratively adjusting the hyperparameters of the intelligent question answering model until the model converges.

[0112] In some embodiments, obtaining real-time user questions through the input module of the exhibition robot includes: collecting user voice signals through the microphone of the input module, converting the voice into text data using speech recognition technology; performing word segmentation, stop word removal, and grammatical error correction on the converted text data, identifying the question type through an intent classification model, and generating structured real-time question information; the question types include at least exhibition process, booth information, job requirements, and event arrangements.

[0113] In some embodiments, the collection of real-time background information of the current user asking the question and real-time dynamic information of the exhibition includes: obtaining the user's on-site registered identity information and job intention through the touch screen or scanning device mounted on the exhibition robot, and generating real-time background information data; obtaining the current time period's booth activity update data and temporarily changed recruitment job demand data from the exhibition real-time management system through the network interface, and storing the exhibition real-time dynamic information in a structured manner according to the timestamp and booth number.

[0114] In some embodiments, the step of inputting the acquired real-time question information, real-time background information, and real-time exhibition dynamic information into a trained intelligent question-answering model to generate personalized answer content for the current user includes: concatenating the semantic vector of the real-time question information, the structured feature vector of the real-time background information, and the timestamp and change feature vector of the real-time exhibition dynamic information as input to the trained intelligent question-answering model; the intelligent question-answering model prioritizes matching recruitment job requirements and booth activity information related to the user's job search intentions based on an attention mechanism to generate initial answer content; post-processing the initial answer content, adjusting the expression according to the user's identity type, embedding temporary changes in the real-time exhibition dynamic information, and forming final answer content that includes the user's personalized needs.

[0115] In some embodiments, the personalized response content generated by the output module of the exhibition robot is fed back to the user in the form of voice or text, including: if voice feedback is selected, the text response content is converted into natural speech through a speech synthesis engine, and the voice parameters are adjusted according to the user's language habits and speech rate preferences; if text feedback is selected, the response content is displayed on the robot's touch screen in a preset format, and interactive booth navigation links or job details links are generated for the user to click and view.

[0116] In some embodiments, the method further includes: during the trade fair, collecting user feedback data on the answers in real time, and incrementally training the intelligent question-answering model through an online learning algorithm; when the frequency of booth activity updates or job posting changes in the real-time dynamic information corresponding to the trade fair exceeds a preset threshold, triggering an adaptive adjustment mechanism for the model parameters, prioritizing the updating of feature weights related to the changed information, so that the intelligent question-answering model outputs real-time adaptations to the dynamic changes at the trade fair.

[0117] In some embodiments, the method further includes: constructing an answer quality assessment model, calculating assessment indicators such as semantic similarity, information completeness, and timeliness scores based on standard answers and actual answer content in historical user question data; after generating personalized answer content, performing real-time verification of the answer content through the answer quality assessment model; if the assessment indicators are lower than preset standards, triggering a candidate answer generation strategy, selecting the candidate answer with the highest matching degree with the user's real-time background information and the real-time dynamic information of the exhibition from a preset answer template library, and replacing the initially generated answer content.

[0118] The embodiments of this application also provide a computer-readable storage medium storing a computer program, the computer program including program instructions, and the processor executing the program instructions to implement the steps of the intelligent question-and-answer method for trade fairs provided in the above embodiments of this application.

[0119] The computer-readable storage medium can be the internal storage unit of the exhibition robot described in the foregoing embodiments, such as the hard drive or memory of the exhibition robot. Alternatively, the computer-readable storage medium can be an external storage device of the exhibition robot, such as a plug-in hard drive, SmartMedia Card (SMC), Secure Digital (SD) card, or Flash Card equipped on the exhibition robot.

[0120] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in this application, and these modifications or substitutions should all be covered within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. An intelligent question-answering method for trade fairs, characterized in that, The method, applied to exhibition robots, includes: The system acquires user question data and corresponding standard answer data from historical job fairs, and obtains participant background information and basic exhibition information associated with the historical question information. The user question data includes historical questions from different types of participants regarding the exhibition process, booth information, job requirements, and event arrangements. The acquired user question data, standard answer data, participant background information, and basic exhibition information are input into the intelligent question answering model for training. By adjusting the parameters of the intelligent question answering model, the corresponding answer content is generated, thus completing the training of the intelligent question answering model. At the trade fair, the input module of the exhibition robot obtains real-time questions from users, collects real-time background information of the current questioning user and real-time dynamic information of the exhibition. The real-time background information includes the user's on-site registered identity information and job intentions, and the real-time dynamic information of the exhibition includes updates on booth activities and temporarily changed recruitment job requirements for the current period. The acquired real-time question information, real-time background information, and real-time exhibition dynamic information are input into a trained intelligent question-answering model to generate personalized answers for the current user. This includes: concatenating the semantic vector of the real-time question information, the structured feature vector of the real-time background information, and the timestamp and change feature vector of the real-time exhibition dynamic information as input to the trained intelligent question-answering model; the intelligent question-answering model prioritizes matching job requirements and booth activity information related to the user's job search intentions based on an attention mechanism to generate initial answer content; post-processing the initial answer content, adjusting the expression according to the user's identity type, and embedding temporary changes from the real-time exhibition dynamic information to form a final answer content that incorporates the user's personalized needs; and feeding back the generated personalized answer content to the user in voice or text form through the exhibition robot's output module.

2. The method according to claim 1, characterized in that, The standard answer data includes standard answers to various types of questions; the participant background information includes the participant's identity type and job search direction; the exhibition basic information includes the exhibition booth distribution, list of recruiting companies, and event schedule; the acquisition of user question data and corresponding standard answer data from historical job fair scenarios, and the acquisition of participant background information and exhibition basic information associated with historical question information, includes: The system extracts user question data and corresponding standard answer data from the user interaction log database of historical job fairs through a data interface. Based on the user ID, it retrieves the corresponding identity type and job search direction from the participant registration database. It also extracts the booth distribution coordinate data, the list of recruiting companies, and the event schedule associated with the booth number and event name involved in the historical questions from the exhibition management system database. The data is then classified and stored according to the question type.

3. The method according to claim 1, characterized in that, The process involves inputting the acquired user question data, standard answer data, participant background information, and basic exhibition information into the intelligent question-answering model for training. By adjusting the parameters of the intelligent question-answering model, corresponding answer content is generated, thus completing the training of the intelligent question-answering model. This includes: The user question data is processed by natural language to extract keywords and semantic vectors. The background information of the participants and the basic information of the exhibition are converted into structured feature vectors. The above vectors are concatenated as the input to the intelligent question answering model. A sequence-to-sequence model or a retrieval-generation hybrid model based on deep learning is adopted as the intelligent question answering model. The standard answer data is used as the target output. The model parameters are optimized through the backpropagation algorithm so that the semantic similarity between the answer content output by the intelligent question answering model and the standard answer reaches a preset threshold. During training, an attention mechanism is incorporated to strengthen the association weight between question keywords and basic exhibition information and attendee background information. The training set and validation set are divided using a cross-validation method, and the hyperparameters of the intelligent question answering model are iteratively adjusted until the model converges.

4. The method according to claim 1, characterized in that, The process of obtaining real-time user questions through the input module of the exhibition robot includes: The input module's microphone collects the user's voice signal, and speech recognition technology is used to convert the speech into text data. The converted text data is processed by word segmentation, stop word removal and grammatical error correction. The question type is identified by the intent classification model and structured real-time question information is generated. The types of questions include at least exhibition process, booth information, job requirements, and event arrangements.

5. The method according to claim 1, characterized in that, The collection of real-time background information of the current user asking the question and real-time dynamic information of the exhibition includes: The system uses touchscreens or QR code scanners mounted on exhibition robots to obtain users' on-site registered identity information and job preferences, generating real-time background information data. The system obtains updated booth activity data and temporarily changed job postings from the exhibition's real-time management system via a network interface, and stores the real-time dynamic information of the exhibition in a structured manner according to timestamps and booth numbers.

6. The method according to claim 1, characterized in that, The personalized responses generated by the exhibition robot's output module are fed back to the user in voice or text form, including: If voice feedback is selected, the text response will be converted into natural speech by a speech synthesis engine, and the voice parameters will be adjusted according to the user's language habits and speaking speed preferences. If the user chooses to provide text feedback, the response will be displayed on the robot's touchscreen in a preset format, along with an interactive booth navigation link or job details link for the user to click and view.

7. The method according to claim 1, characterized in that, The method further includes: During the trade fair, user feedback data on the answers is collected in real time, and the intelligent question-answering model is incrementally trained through online learning algorithms. When the frequency of updates to booth activities or changes in job postings in the real-time dynamic information of the trade fair exceeds a preset threshold, the adaptive adjustment mechanism of the model parameters is triggered. The feature weights related to the changed information are updated first, so that the output of the intelligent question answering model can adapt to the dynamic changes at the trade fair in real time.

8. The method according to claim 1, characterized in that, The method further includes: A response quality assessment model is constructed, which calculates assessment indicators such as semantic similarity, information completeness, and timeliness scores based on standard answers and actual response content in historical user question data. After generating personalized answer content, the answer quality assessment model is used to verify the answer content in real time. If the assessment index is lower than the preset standard, the alternative answer generation strategy is triggered. The alternative answer with the highest matching degree with the user's real-time background information and the real-time dynamic information of the exhibition is selected from the preset answer template library and replaced with the initially generated answer content.

9. A convention and exhibition robot, characterized in that, The exhibition robot includes a memory and a processor; The memory is used to store computer programs; The processor is configured to execute the computer program and, in executing the computer program, implement the method as described in any one of claims 1 to 8.

Images