Information detection method and system, and electronic device

Abstract

Disclosed in the present disclosure are an information detection method and system, and an electronic device. The method comprises: detecting input information; inputting the input information into a dialogue model for analysis to obtain at least one piece of reply information matching the input information; acquiring a target text feature of the reply information from internal state information of the dialogue model, wherein the internal state information is used for representing a rule for analyzing the input information by the dialogue model, and the target text feature is used for representing the semantics of the reply information; and determining a detection result of the reply information on the basis of the target text feature of the reply information, wherein the detection result is used for indicating whether the reply information that is output by the dialogue model is reliable.

Description

[0001] Information Detection Method, System, and Electronic Device Cross-Reference This disclosure claims priority to Chinese patent application number 202410099131.3, filed with the Patent Office of China on January 23, 2024, entitled "Information Detection Method, System, and Electronic Device," the entire contents of which are incorporated herein by reference. Technical Field This disclosure relates to the fields of large model technology and large language model detection technology, and more specifically, to an information detection method, system, and electronic device. Background: With the rapid development of large models, the requirements for the authenticity and reliability of the response information output by large models are also increasing. Detecting the reliability of the response information output by large models has become increasingly important. In related technologies, hallucination detection methods based on preplexity or self-checking (GPT) are commonly used to determine the reliability of the response information output by large models. Among them, the preplexity-based hallucination detection method primarily determines the uncertainty of each word in the reply information output by the large model, and then multiplies the uncertainty of each word to obtain the uncertainty of the reply information output by the large model. However, due to the inaccurate estimation of the uncertainty of the sentence length of the reply information output by the large model and the diversity of the expression forms of the sentences in the reply information output by the large model, the reliability detection results of the reply information output by the large model may be inaccurate. Therefore, the above method has the technical problem of being unable to effectively detect the reply information. To address this problem, no effective solution has been proposed. SUMMARY OF THE INVENTION Embodiments of the present disclosure provide an information detection method, system, and electronic device to at least address the technical problem of being unable to effectively detect the reliability of reply information. According to one aspect of an embodiment of the present disclosure, a method for detecting information is provided. The method may include: monitoring input information; inputting the input information into a dialogue model for analysis to obtain at least one reply matching the input information; obtaining target text features of the reply from internal state information of the dialogue model, wherein the internal state information represents the rules used by the dialogue model to analyze the input information, and the target text features represent the semantics of the reply; and determining a reply detection result based on the target text features of the reply, wherein the detection result indicates whether the reply output by the dialogue model is reliable. According to another aspect of an embodiment of the present disclosure, a method for generating information is also provided, which is applied to a question-answering system deployed in a scenario task.The method may include: monitoring input information in a scenario task on an operation interface of a question-answering system; calling a dialogue model that matches the scenario task, inputting the input information into the dialogue model for analysis, and obtaining at least one reply information that matches the input information; obtaining target text features of the reply information from internal state information of the dialogue model, wherein the internal state information is used to represent the rules for the dialogue model to analyze the input information, and the target text features are used to represent the semantics of the reply information in the scenario task; determining a detection result of the reply information based on the target text features of the reply information; in response to the detection result that the reply information output by the dialogue model is reliable in the scenario task, outputting the reply information; in response to the detection result that the reply information output by the dialogue model is unreliable in the scenario task, outputting corresponding prompt information. According to another aspect of an embodiment of the present disclosure, a method for detecting information is also provided, which may include: monitoring input information on a dialogue interface; displaying at least one reply information matching the input information on the dialogue interface, wherein the reply information is obtained by analyzing the input information using a dialogue model; in response to an information detection operation performed on the dialogue interface, displaying a detection result of the reply information on the dialogue interface, wherein the detection result is used to indicate whether the reply information output by the dialogue model is reliable, and is determined based on target text features of the reply information, wherein the target text features are used to represent the semantics of the reply information and are obtained from internal state information of the dialogue model, and the internal state information is used to represent the rules used by the dialogue model to analyze the input information. According to one aspect of an embodiment of the present disclosure, a device for detecting information is provided, which may include: a first monitoring component, configured to monitor input information; a first analyzing component, configured to input the input information into a dialogue model for analysis to obtain at least one reply information matching the input information; a first acquiring component, configured to acquire target text features of the reply information from internal state information of the dialogue model, wherein the internal state information is used to represent the rules for the dialogue model to analyze the input information, and the target text features are used to represent the semantics of the reply information; a first determining component, configured to determine a detection result of the reply information based on the target text features of the reply information, wherein the detection result is used to indicate whether the reply information output by the dialogue model is reliable.According to another aspect of an embodiment of the present disclosure, there is also provided an information generation device, which may include: a second monitoring component, configured to monitor input information in a scenario task on an operation interface of a question-answering system; a second analysis component, configured to call a dialogue model that matches the scenario task, input the input information into the dialogue model for analysis, and obtain at least one reply information that matches the input information; a second acquisition component, configured to obtain target text features of the reply information from internal state information of the dialogue model, wherein the internal state information is used to represent the rules for the dialogue model to analyze the input information, and the target text features are used to represent the semantics of the reply information in the scenario task; a second determination component, configured to determine a detection result of the reply information based on the target text features of the reply information; a first output component, configured to output reply information in response to a detection result that the reply information output by the dialogue model is reliable in the scenario task; and a second output component, configured to output corresponding prompt information in response to a detection result that the reply information output by the dialogue model is unreliable in the scenario task. According to another aspect of an embodiment of the present disclosure, an information detection device is further provided, which may include: a third monitoring component, configured to monitor input information on a dialogue interface; a first display component, configured to display at least one reply information matching the input information on the dialogue interface, wherein the reply information is obtained by analyzing the input information using internal state information of a dialogue model; a second display component, configured to respond to an information detection operation performed on the dialogue interface, and display a detection result of the reply information on the dialogue interface, wherein the detection result is used to indicate whether the reply information output by the dialogue model is reliable, and is determined based on target text features of the reply information, the target text features being used to indicate the semantics of the reply information and being obtained from the internal state information of the dialogue model, and the internal state information being used to indicate the rules by which the dialogue model analyzes the input information.According to another aspect of an embodiment of the present disclosure, an information detection system is provided. The system may include: an information input terminal configured to monitor input information; an information detection terminal configured to input the input information into a dialogue model for analysis to obtain at least one reply message matching the input information; obtaining target text features of the reply message from internal state information of the dialogue model, wherein the internal state information represents the rules used by the dialogue model to analyze the input information, and the target text features represent the semantics of the reply message; determining a reply message detection result based on the target text features of the reply message, wherein the detection result indicates whether the reply message output by the dialogue model is reliable; an information output terminal configured to output the reply message in response to a detection result indicating that the reply message output by the dialogue model is reliable; and output corresponding prompt information in response to a detection result indicating that the reply message output by the dialogue model is unreliable. According to another aspect of an embodiment of the present disclosure, an electronic device is provided, including a memory and a processor, wherein the memory is configured to store computer-executable instructions, and the processor is configured to execute the computer-executable instructions, wherein the computer-executable instructions are executed by the processor to perform the steps of the information detection method. According to another aspect of an embodiment of the present disclosure, a computer-readable storage medium is provided. The computer-readable storage medium includes a stored program. When the program is executed by a processor, the program controls the device containing the computer storage medium to execute the steps of the information detection method. According to another aspect of an embodiment of the present disclosure, a computer program is provided. When the computer program is executed by a processor, the computer program implements the information detection method in each embodiment of the present disclosure. According to another aspect of an embodiment of the present disclosure, a computer program product is provided. The computer program product includes a non-volatile computer-readable storage medium. The non-volatile computer-readable storage medium stores the computer program. When the computer program is executed by a processor, the computer program implements the information detection method in each embodiment of the present disclosure. According to another aspect of an embodiment of the present disclosure, a computer program product is provided. When the computer program product is executed by a processor, the computer program implements the information detection method in each embodiment of the present disclosure. In an embodiment of the present disclosure, input information is monitored; the input information is input into a dialogue model for analysis to obtain at least one piece of reply information that matches the input information; target text features of the reply information are obtained from internal state information of the dialogue model, wherein the internal state information is used to represent the rules used by the dialogue model to analyze the input information, and the target text features are used to represent the semantics of the reply information; and based on the target text features of the reply information, a detection result of the reply information is determined, wherein the detection result is used to indicate whether the reply information output by the dialogue model is reliable.That is, in the present disclosure, input information is analyzed based on a dialogue model to obtain at least one reply matching the input information. Since the dialogue model's internal state information represents the rules by which the dialogue model analyzes the input information, the target text features of the reply information can be obtained from the dialogue model's internal state information, thereby better mining and utilizing the semantic features of the reply information. The target text features are used to determine a reply information detection result. This detection result can characterize the overall uncertainty of the reply information, that is, reflect the reliability of the reply information output by the dialogue model. This effectively detects the reliability of the reply information output by the dialogue model, thereby resolving the technical problem of being unable to effectively detect the reliability of the reply information. It should be noted that the general description above and the detailed description that follow are merely illustrative and illustrative of the present disclosure and do not constitute a limitation of the present disclosure. BRIEF DESCRIPTION OF THE DRAWINGS The drawings described herein are provided to provide a further understanding of the present disclosure and constitute a part of the present disclosure. The illustrative embodiments of the present disclosure and their descriptions are provided to explain the present disclosure and do not constitute an undue limitation of the present disclosure. In the accompanying drawings: FIG1 is a schematic diagram of an application scenario of an information detection method according to an embodiment of the present disclosure; FIG2 is a flow chart of an information detection method according to an embodiment of the present disclosure; FIG3 is a flow chart of another information generation method according to an embodiment of the present disclosure; FIG4 is a flow chart of another information detection method according to an embodiment of the present disclosure; FIG5 is a schematic diagram of an information detection system according to an embodiment of the present disclosure; FIG6 is a flow chart of another information detection method according to an embodiment of the present disclosure; FIG7 is a schematic diagram of a feature response of a text feature set according to an embodiment of the present disclosure; FIG8 is a schematic diagram of a large model knowledge hallucination detection according to an embodiment of the present disclosure; FIG9 is a schematic diagram of an information detection device according to an embodiment of the present disclosure; FIG10 is a schematic diagram of another information generation device according to an embodiment of the present disclosure; FIG11 is a schematic diagram of another information detection device according to an embodiment of the present disclosure; FIG12 is a structural block diagram of a computer terminal according to an embodiment of the present disclosure. DETAILED DESCRIPTION OF THE EMBODIMENTS To help those skilled in the art better understand the present disclosure, the technical solutions in the embodiments of the present disclosure will be clearly and completely described below in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present disclosure, not all of them. Based on the embodiments of the present disclosure, all other embodiments obtained by persons of ordinary skill in the art without creative work shall fall within the scope of protection of the present disclosure.It should be noted that the terms "first," "second," and so on, in the specification and claims of this disclosure, and in the accompanying drawings, are used to distinguish similar objects and are not necessarily used to describe a specific order or precedence. It should be understood that such terms are interchangeable where appropriate, such that the embodiments of the disclosure described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "including," "having," and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product, or apparatus comprising a series of steps or components is not necessarily limited to those steps or components expressly listed, but may include other steps or components not expressly listed or inherent to such process, method, product, or apparatus. The technical solutions provided in this disclosure are primarily implemented using large-scale model technology. Large-scale models herein refer to deep learning models with large-scale model parameters, typically including hundreds of millions, tens of billions, hundreds of billions, trillions, or even more than ten trillion model parameters. Large models, also known as cornerstone models or foundation models, are pre-trained on large, unlabeled corpora, producing pre-trained models with over 100 million parameters. These models are adaptable to a wide range of downstream tasks and exhibit good generalization capabilities. Examples include large language models (LLMs) and multi-modal pre-training models. It should be noted that in practical applications, large models can be fine-tuned using a small number of samples, allowing them to be applied to different tasks. For example, large models are widely applicable in fields such as natural language processing (NLP), computer vision, and speech processing. Specifically, they can be applied to computer vision tasks such as visual question answering (VQA), image captioning (IC), and image generation. They can also be widely used in natural language processing tasks such as text-based sentiment classification, text summarization, and machine translation. Therefore, the main application scenarios of large models include but are not limited to digital assistants, intelligent robots, search, online education, office software, e-commerce, intelligent design, etc.In the embodiments of the present disclosure, data processing using a large language model in a knowledge hallucination detection scenario is used as an example for explanation. First, some nouns or terms that appear in the description of the embodiments of the present disclosure are applicable to the following explanations: A large language model, also known as a large-scale language model, is a type of language model with a very large number of parameters (for example, more than 1B), represented by a Transformer model, and has good natural language generation capabilities. Knowledge hallucination refers to the situation where the output of a large model is "inconsistent with the facts" or "made out of thin air." Internal states of the large model refer to information related to weights and features within the large model. Feature clipping (FC) refers to clipping abnormal features in the hidden layer of the model. Token embedding refers to the fact that different words have a corresponding feature representation in the output layer. Perplexity is a measure of output confidence calculated based on the logical probability (logit) of the large model output. According to an embodiment of the present disclosure, a method for detecting information is provided. It should be noted that the steps illustrated in the flowcharts of the accompanying drawings can be executed in a computer system, such as a set of computer-executable instructions. Furthermore, although the flowcharts illustrate a logical order, in some cases, the steps illustrated or described may be executed in a different order. Considering the large number of model parameters in a large model and the limited computing resources of a mobile terminal, the information detection method provided in the embodiments of the present disclosure can be applied to, but is not limited to, the application scenario shown in FIG1 . FIG1 is a schematic diagram of an application scenario of an information detection method according to an embodiment of the present disclosure. In the application scenario shown in FIG1 , a large model is deployed on a server 10. The server 10 can be connected to one or more client devices 20 via a local area network, a wide area network, the Internet, or other types of data networks. Client devices 20 may include, but are not limited to, smartphones, tablet computers, laptops, PDAs, personal computers, smart home devices, and in-vehicle devices. Client devices 20 can interact with users via a graphical user interface to invoke the large model, thereby implementing the method provided in the embodiments of the present disclosure. In an embodiment of the present disclosure, a system composed of a client device and a server may perform the following steps: The client device executes input information, and the input information may be a type of inquiry information, query information, etc.The server executes step S101 to monitor input information. Step S102 inputs the input information into the dialogue model for analysis to obtain at least one reply matching the input information. Step S103 obtains target text features of the reply information from the internal state information of the dialogue model. Step S104 determines a reply information detection result based on the target text features of the reply information, where the detection result indicates whether the reply information output by the dialogue model is reliable. It should be noted that if the client device's operating resources meet the deployment and operating requirements of a large model, the embodiments of the present disclosure can be implemented on the client device, and the comparison model used in the present disclosure can be a generative dialogue model. In this operating environment, the present disclosure provides an information detection method as shown in Figure 2. Figure 2 is a flow chart of an information detection method according to an embodiment of the present disclosure. As shown in Figure 2, the method may include the following steps: Step S201: Monitoring input information. In the technical solution provided in step S201 of the present disclosure, the input information may be multimodal information. The types of multimodal information include at least one of the following: text information containing character information, video frame information containing frame image information, and audio information. For example, this may be a query or a chat message, without specific limitation. The question-and-answer system can monitor the input information. For example, if the input information is text information, a user can enter multiple text questions into the question-and-answer system. These multiple text questions constitute the input information. The question-and-answer system can monitor the multiple questions in real time and generate corresponding response information based on the multiple questions. It should be noted that when the input information is non-text information, such as video or audio information, the video or audio information can be converted into text information for processing. In step S202, the input information is input into a dialogue model for analysis to obtain at least one response information that matches the input information. In the technical solution provided in step S202 of the present disclosure, the dialogue model can be a large model capable of understanding and generating natural language text, such as a generative dialogue model. For example, the dialogue model is a model pre-trained using natural language processing technology and machine learning algorithms, and is used to generate dialogue content, answer questions, provide suggestions, etc. Based on this, after monitoring input information in step S201, the input information can be input into the dialogue model for analysis, thereby obtaining at least one reply information that matches the input information. The reply information can be multimodal information, and the reply information type can include at least one of the following: text information, image information, video information, and voice information.In this embodiment, the dialogue model is used to analyze user input information to generate at least one reply matching the input information. Therefore, after the question-answering system detects input information, it can directly input the input information into the dialogue model, where it can analyze the input information and generate at least one reply matching the input information. Step S203: Target text features of the reply information are obtained from the dialogue model's internal state information. In the technical solution provided in step S203 of the present disclosure, after obtaining at least one reply matching the input information in step S202, the target text features of the reply information can be obtained from the dialogue model's internal state information. The internal state information represents the rules used by the dialogue model to analyze the input information, and the target text features represent the semantics of the reply information. For example, the target text features can be features of the sentence corresponding to the reply information. In this embodiment, the dialogue model's internal state information can include model parameters, features, weights, learning rules, and so on. This internal state information can influence the dialogue model's processing of input information and the processing results. When processing input information, the dialogue model can analyze and infer the input information based on its internal state information, thereby generating corresponding output results. For example, the dialogue model's internal state information may include the rules used by multiple network layers within the dialogue model to analyze the input information. The multiple network layers of the dialogue model may include decoder layers, fully connected layers (FC layers), etc. These multiple network layers are used to perform different processing operations on the input information to generate output results. The target text features of the response information can be obtained from the output results of these multiple network layers. For example, after processing the input information, the multiple network layers in the dialogue model can generate response information that matches the input information. The response information may include multiple characters, each of which corresponds to a feature vector in different network layers of the dialogue model. Since in the large model, the features of the last character in the generated reply information can often represent the features of the entire reply information, based on this, the feature vector corresponding to the last character in the reply information in the middle layer of multiple network layers of the dialogue model can be determined according to the arrangement order of multiple characters in the reply information, and then the middle layer feature vector corresponding to the last character of the multiple characters in the reply information can be determined as the target text feature of the reply information.Taking the Large Language Model Meta-7B (llama-7B) as an example, assuming the llama-7B model includes 33 network layers, after the llama-7B model processes the input information, the feature of the last character output by the 17th network layer (the middle layer) can be selected as the target text feature of the reply information output by the llama-7B model. In step S204, a detection result of the reply information is determined based on the target text feature of the reply information. In the technical solution provided in step S204 of the present disclosure, after obtaining the target text feature of the reply information in step S203, a detection result of the reply information can be determined based on the target text feature of the reply information. The detection result indicates whether the reply information output by the dialogue model is reliable. In this embodiment, after obtaining the target text feature of the reply information, a covariance matrix can be constructed based on the target text feature of the reply information. A metric for the reply information output by the dialogue model can then be calculated based on the covariance matrix. The metric can be a hallucination degree metric score corresponding to the reply information output by the dialogue model. After obtaining the metric, the reliability of the response information output by the dialogue model can be further determined based on the metric and the metric threshold. For example, assuming the metric is the hallucination degree metric score corresponding to the response information output by the dialogue model, the metric threshold can be a score threshold. In this case, the hallucination degree metric score can be compared with the score threshold. When the hallucination degree metric score is greater than the score threshold, it indicates that the output of the dialogue model contains hallucinations, that is, the response information output by the dialogue model is unreliable. When the hallucination degree metric score is not greater than the score threshold, it indicates that the output of the dialogue model does not contain hallucinations, that is, the response information output by the dialogue model is reliable. Based on steps S201 to S204 of the above embodiment, input information is analyzed based on the dialogue model to obtain at least one reply matching the input information. Since the dialogue model's internal state information represents the rules used by the dialogue model to analyze the input information, the target text features of the reply information can be obtained from the dialogue model's internal state information. This allows for better mining and utilization of the semantic features of the reply information. The target text features are used to determine a reply detection result. This detection result can represent the overall uncertainty of the reply information, specifically, whether the reply information output by the dialogue model is reliable. This effectively detects the reliability of the reply information output by the dialogue model, thereby resolving the technical issue of being unable to effectively detect the reply information output by the model. The above-described method of this embodiment is further described below.As an optional implementation, step S203, obtaining target text features corresponding to the reply information from the internal state information of the dialogue model, includes: obtaining a text feature set of the reply information in the network layer of the dialogue model, wherein the internal state information includes a text feature set in the network layer, and the text feature set in the network layer includes text features of text units constituting the reply information; and determining the target text features from the text feature set in the network layer. In this embodiment, the dialogue model may include multiple network layers, each of which may output text features of each text unit in the reply information. The text features of each text unit may constitute the text feature set of the reply information in the network layer of the dialogue model. Based on this, the text feature set of the reply information in the network layer of the dialogue model may be obtained. For example, the text feature set of the reply information in the network layer of the dialogue model can be represented as a token embedding, where the text units of the text features included in the text feature set can be represented as tokens. By obtaining the text features of each text unit in the reply information output by each network layer of the dialogue model and then combining the text features of each text unit, the text feature set of the reply information in each network layer of the dialogue model can be obtained. In this embodiment, the text feature set is directly extracted from the reply information output by the network layer of the dialogue model, eliminating the need to use an additional model to extract the text features of the reply information output by the dialogue model, thereby reducing computational overhead and improving computational efficiency. Optionally, after determining the text feature set of the reply information in each network layer of the dialogue model, a target text feature can be determined from the text feature set of the network layer. As an optional implementation, determining the target text feature from the text feature set in the network layer includes: determining, from the text feature set in the network layer, a text feature corresponding to a target text unit in the reply information, where the target text unit includes the semantics of the reply information; and determining the text feature corresponding to the target text unit as the target text feature. In this embodiment, since the text feature set includes the text features of each text unit that constitutes the reply message, and the target text unit includes the semantics of the reply message, the text features corresponding to the target text unit in the reply message can be determined from the network-level text feature set, and the text features corresponding to the target text unit can be determined as the target text features. For example, in the large model, the text features of the last character of a sentence often include the semantic information of the entire sentence. Based on this, the last text unit in the reply message can be used as the target text unit for the reply message. In other words, the last character of the sentence corresponding to the reply message can be used as the target text unit for the reply message.The text feature corresponding to the last text unit in the reply message can be determined from the text feature set in the network layer according to the order of the text units in the reply message. The text feature corresponding to the last text unit can then be determined as the text feature corresponding to the target text unit in the reply message. As an optional implementation, obtaining the text feature set of the reply message in the network layer of the dialogue model includes: obtaining the text feature set of the reply message in the output network layer of the dialogue model; and determining, in the target hidden network layer of the dialogue model, the text feature set of the target hidden network layer that corresponds to the text feature set of the output network layer. In this embodiment, the network layer of the dialogue model may include an output network layer and a target hidden network layer, wherein the output network layer may be the penultimate layer in the network layer of the dialogue model, and the target hidden network layer may be an intermediate layer in the network layer of the dialogue model. For example, as can be seen from the foregoing description, a text feature set is composed of the text features of each text unit constituting the reply information. Based on this, the text features of each text unit output by the output network layer of the dialogue model can be obtained, and then the text features of each text unit can be combined to form the text feature set of the output network layer of the dialogue model. This text feature set can be called a token embedding. Optionally, since the target hidden network layer of the dialogue model can be an intermediate layer among multiple network layers of the dialogue model, based on this, a text feature set of the target hidden network layer corresponding to the text feature set of the output network layer can be determined in the target hidden network layer of the dialogue model. The process of determining the text feature set of the target hidden network layer corresponding to the text feature set of the output network layer in the target hidden network layer of the dialogue model is further described below. As an optional implementation, determining the text feature set of the target hidden network layer corresponding to the text feature set of the output network layer in the target hidden network layer of the dialogue model includes: updating the text feature set of the output network layer to obtain an updated text feature set, wherein the updated text feature set does not include noise features; and, in the target hidden network layer, The text feature set of the target hidden network layer corresponding to the updated text feature set is determined. In this embodiment, the text features included in the text feature set of the output network layer may contain a large number of extreme feature responses, which can be considered noise features. For example, some text features included in the text feature set may have excessively large or small response amplitudes. Such text features can easily cause the dialogue model to output erroneous classification results. Based on this, after obtaining the text feature set of the dialogue model's output network layer, the text feature set can be updated to obtain an updated text feature set.For example, the text feature set of the output network layer can be updated by adjusting the response amplitudes of the text features included in the text feature set. For example, the text features in the text feature set can be adjusted to fall within the response amplitude threshold range corresponding to the text feature set of the output network layer, thereby updating the text feature set in the output network layer. Alternatively, the text feature set can be updated by performing a feature clipping operation on the text feature set of the output network layer. For example, text features with excessively large or small response amplitudes in the text feature set can be clipped to obtain a clipped text feature set, which is then determined as the updated text feature set to remove noise features from the text feature set. Clipping the text feature set is merely an illustrative example of updating a text feature set and does not limit the method for updating the text feature set. In this embodiment, updating the text feature set of the output network layer can remove noise features from the text feature set, thereby improving the output quality of the dialogue model and avoiding hallucination outputs with high consistency due to oversaturated classification results output by the dialogue model. Next, the process of updating the text feature set of the output network layer is further described. As an optional implementation, updating the text feature set of the output network layer to obtain an updated text feature set includes: in response to a response amplitude of a text feature in the text feature set of the output network layer being less than a first response amplitude threshold, adjusting the response amplitude to the first response amplitude threshold; in response to a response amplitude of a text feature in the text feature set of the output network layer being greater than or equal to the first response amplitude threshold and less than or equal to a second response amplitude threshold, maintaining the response amplitude, wherein the second response amplitude threshold is greater than or equal to the first response amplitude threshold; in response to a response amplitude of a text feature in the text feature set of the output network layer being greater than the second response amplitude threshold, adjusting the response amplitude to the second response amplitude threshold; and determining the response amplitude adjusted to the first response amplitude threshold, the maintained response amplitude, and the response amplitude adjusted to the first response amplitude threshold as the updated text feature set. In this embodiment, the first response amplitude threshold and the second response amplitude threshold are used to measure the magnitude of the response amplitudes of the text features in the text feature set. For example, the first response amplitude threshold can represent the minimum response amplitude of the text feature, and the second response amplitude threshold can represent the maximum response amplitude of the text feature. Based on this, the response amplitude of each text feature included in the text feature set of the output network layer is compared with the first response amplitude threshold and the second response amplitude threshold, respectively, to determine the response amplitude to be adjusted in the text feature set.If a text feature in the text feature set of the output network layer has a response amplitude less than the first response amplitude threshold, it indicates that the response amplitude of the text feature is too low. In this case, the response amplitude of the text feature in the text feature set can be adjusted to the first response amplitude threshold. Similarly, if a text feature in the text feature set of the output network layer has a response amplitude greater than the second response amplitude threshold, it indicates that the response amplitude of the text feature is too high. In this case, the response amplitude of the text feature in the text feature set can be adjusted to the second corresponding amplitude threshold. If a text feature in the text feature set of the output network layer is greater than or equal to the first response amplitude threshold and less than or equal to the second response amplitude threshold, it indicates that the response amplitude of the text feature in the text feature set is within a normal range. In this case, the response amplitude of the text feature in the text feature set can be left unprocessed. After determining the first and second response amplitude thresholds corresponding to the text feature set of the output network layer, the text features within the range of the first and second response amplitude thresholds can be determined as the text features included in the updated text feature set. In other words, all text features included in the updated text feature set are within the response amplitude threshold range. For example, a piecewise function can be used to determine whether the response amplitudes of text features in a text feature set are within a normal range, and to remove text features whose response amplitudes are less than a first response amplitude threshold and whose response amplitudes are greater than a second response amplitude threshold. The piecewise function is shown in the following formula (1): hmin, h < h. min , h min < h < h max ( 1 ) hmax' h > h maxHere, FC(h) can be used to represent a piecewise function; hmin can be used to represent the first response amplitude threshold; h can be used to represent the response amplitude of text features within the text feature set that are within the normal range; and hmax can be used to represent the second response amplitude threshold. As can be seen from the above formula, the response amplitude of a text feature in the text feature set is greater than or equal to the first response amplitude threshold and less than or equal to the second response amplitude threshold. Based on this, when the response amplitude h of a text feature in the text feature set is less than the first response amplitude threshold hmin, the response amplitude can be truncated to the first response amplitude threshold hmin; when the response amplitude of a text feature in the text feature set is greater than the second response amplitude threshold hmax, the response amplitude can be truncated to the second response amplitude threshold hmax to ensure that the response amplitudes of all text features in the text feature set are within the response amplitude threshold range. Optionally, the first response amplitude threshold hmin and the second response amplitude threshold hmax can be obtained using dynamically stored feature responses in a memory bank. MemoryBank is a first-in, first-out feature memory capable of storing feature responses for 3,000 text units. The feature responses for these 3,000 text units are sorted, with hmax set to the 0.2% percentile of the maximum response amplitude, and hmin set to the 0-2% percentile of the minimum response amplitude. This is merely an example, and the method for determining the first response amplitude threshold hmin and the second response amplitude threshold hmax is not limited. As an optional implementation, the reply information detection method further includes: obtaining the order of arrangement of the multiple hidden network layers in the dialogue model; and determining a target hidden network layer from the multiple hidden network layers based on the order of arrangement. In this embodiment, the dialogue model includes multiple hidden network layers, which help the production dialogue model learn more complex semantic and grammatical rules, thereby generating more accurate responses. The multiple hidden network layers have a corresponding order in the dialogue model. Based on this, the order of the multiple hidden network layers in the dialogue model can be determined, and then a target hidden network layer can be determined from the multiple hidden network layers based on the order. The target hidden network layer can be a middle layer among the multiple hidden network layers. This is merely an example and does not limit the specific position of the target hidden network layer in the dialogue model. The following further describes the process of determining the target hidden network layer from the multiple hidden network layers based on the order of the multiple hidden network layers. As an optional embodiment, determining the target hidden network layer from the multiple hidden network layers based on the order of the multiple hidden network layers includes determining the hidden network layer in the middle of the order as the target hidden network layer.In this embodiment, after determining the order of the multiple hidden network layers in the dialogue model, the hidden network layer in the middle of the order can be determined as the target hidden network layer. That is, the target hidden network layer can be the middle layer among the multiple hidden network layers. For example, assuming the dialogue model is the llama-7B model and that the llama-7B model contains 33 hidden network layers, the 17th hidden network layer (the middle layer) can be selected as the target hidden network layer. As an optional implementation, step S204, determining the detection result of the reply information based on the target text features of the reply information, includes: determining a metric for the reply information based on the target text features of the reply information, where the metric is used to indicate the reliability of the reply information output by the dialogue model; and determining the detection result of the reply information based on the metric. In this embodiment, after determining the target text features of the reply information, a feature covariance matrix for the reply information can be constructed based on the target text features of the reply information. Furthermore, a metric for the reply information can be determined based on the feature covariance matrix, where the metric is used to indicate the reliability of the reply information output by the dialogue model. For example, the metric can be a hallucination degree metric score of the dialogue model. This metric can better measure the inconsistency of the response information output by the dialogue model. This is merely an example and does not limit the specific content of the metric. For example, assuming the metric is a hallucination degree metric score, in this case, after obtaining the feature covariance matrix of the response information, the metric can be calculated using the following formula (2).

[0002] E = -logdet( Z + aI K ) ( 2)

[0003] K Where E can be used to represent a metric, K can be used to represent the number of response messages output by multiple network layers of the conversation model, £ can be used to represent the feature covariance matrix, OCIK is a small regularization term used to prevent the covariance matrix from being a non-full rank matrix, IK can be used to represent the identity matrix with latitude K, and a is a characteristic parameter, which can be 0.001. Optionally, since the determinant of a matrix can be obtained by calculating the eigenvalue, the metric in the above formula (2) can be expressed by the following formula (3):

[0004] E = mog(n& = jlX」og0i) (3) Where, λ={λ], λ2, λ refers to the matrix S + « I KK feature values ​​of the reply information. Optionally, after determining the metric for the reply information, a detection result for the reply information can be determined based on the metric, where the detection result can be used to indicate whether the reply information output by the dialogue model is reliable. Optionally, after determining the target text features of the reply information, the target text features of the reply information can be evaluated using a machine learning model to determine the detection result for the reply information, or a text similarity algorithm can be used to measure the similarity between the reply information and the input information to determine the reliability of the reply information. This is merely an example and does not limit the specific method for determining the detection result for the reply information based on the target text features of the reply information. As an optional embodiment, determining the detection result for the reply information based on the metric includes: in response to the metric being greater than a metric threshold, determining the detection result as unreliable; and in response to the metric being less than or equal to the metric threshold, determining the detection result as reliable. In this embodiment, because the reply information output by the dialogue model suffers from knowledge hallucination, i.e., the reply information output by the dialogue model is unreliable, a metric threshold can be used to measure the metric of the reply information output by the dialogue model to determine whether the reply information output by the dialogue model is reliable. For example, the metric of the reply information output by the dialogue model can be compared with the metric threshold, and the reliability of the reply information output by the dialogue model can be determined based on the comparison result. For example, if the metric is greater than the metric threshold, it indicates that the reply information output by the dialogue model may have a high degree of hallucination. Based on this, the reply information detection result can be determined as unreliable. Conversely, if the metric is less than or equal to the metric threshold, it indicates that the reply information output by the dialogue model has a low degree of hallucination. In this case, the reply information detection result can be determined as reliable. As an optional implementation, the reply information detection method further includes: if the detection result indicates that the reply information output by the dialogue model is unreliable, prohibiting the output of the reply information; and if the detection result indicates that the reply information output by the dialogue model is reliable, outputting the reply information. In this embodiment, when the response information output by the dialogue model is unreliable, the dialogue model may be prohibited from outputting the response information. When the response information output by the dialogue model is reliable, the dialogue model is allowed to output the response information. This ensures that only response information that has been verified to be reliable is output, thereby preventing misleading or erroneous information from being disseminated.As an optional implementation, determining a metric for the reply information based on the target text features of the reply information includes: constructing a feature covariance matrix using at least one target text feature of at least one reply information; and determining the metric for the reply information based on the feature covariance matrix. In this embodiment, as described above, the target text features are text features corresponding to the target text units of the reply information, and the target text units include the semantics of the reply information. Based on this, when determining the metric for the reply information based on the target text features of the reply information, a feature covariance matrix can be constructed based on the target text features of the reply information, and the metric for the reply information can be determined based on the feature covariance matrix. For example, when the at least one reply information includes multiple reply information, since each of the multiple reply information corresponds to a target text feature, the feature covariance matrix can be constructed using the following formula (4).

[0005] 1 = Z T -J d - Z (4) Wherein, Z can be used to represent the target text feature matrix corresponding to the reply information, ZT can be used to represent the transposed matrix of the target text feature matrix corresponding to the reply information, Jd can be used to represent the centralization matrix, where, Jd = Id IK IK, where, can be used to represent a full-one column vector with latitude K. Optionally, the target text feature matrix Z corresponding to the reply information can be obtained by concatenating the target text features corresponding to multiple reply information. For example, Z = [Z1, Z2, ... Zj ... , Z k ], where Zi can be used to represent the target text feature hT corresponding to the i-th reply information, that is, Zj = h T, where the target text feature station can be the text feature of the last text unit of the i-th reply information. Optionally, when the at least one reply information includes only one reply message, K in the above formula can be set to 1, that is, K=1. Then, a covariance matrix is ​​constructed according to the method described above, and the covariance matrix is ​​used to determine the metric of the reply information. This will not be described in detail here. As an optional implementation, determining the metric of the reply information based on the feature covariance matrix includes: determining the logarithmic determinant of the feature covariance matrix; and determining the metric corresponding to the logarithmic determinant. In this embodiment, since the feature covariance matrix is ​​a matrix used to describe the correlation and variance between different features, after determining the feature covariance matrix, the logarithmic determinant of the feature covariance matrix can be further determined. The logarithmic determinant can be used to measure the degree of correlation and variance between features. Based on the logarithmic determinant, a metric corresponding to the logarithmic determinant is determined to measure the reliability of the reply information output by the model. In the above steps, by calculating the metrics of the response information output by the dialogue model and then directly using the metrics to determine the reliability of the response information output by the dialogue model, the uncertainty of the entire output sentence corresponding to the response information can be better characterized, thereby reflecting the degree of knowledge illusion of the response information, that is, the reliability of the response information. Under the above operating environment, the present disclosure also provides an information generation method as shown in Figure 3, which is applied to a question-and-answer system deployed in a scenario task, such as a generative question-and-answer system. Figure 3 is a flowchart of another information generation method according to an embodiment of the present disclosure. As shown in Figure 3, the method may include the following steps. Step S301: Monitoring input information in the scenario task on the operation interface of the question-and-answer system. In the technical solution provided in step S301 above, the question-and-answer system can be deployed in the scenario task, and the question-and-answer system can generate corresponding response information based on the input information provided by the user. The operation interface can be an interface for the user to interact with the question-and-answer system. The user can enter input information related to the scenario task on the operation interface, where the input information can be any question posed by the user. The input information in the scenario task can be monitored on the operation interface of the question-and-answer system. Step S302: retrieve a dialogue model that matches the scenario task, input the input information into the dialogue model for analysis, and obtain at least one reply information that matches the input information.In the technical solution provided in step S302 of the present disclosure, after detecting input information in a scenario task, a dialog model matching the scenario task can be retrieved. The dialog model can be a large model. The input information is then fed into the dialog model for analysis to obtain at least one response message matching the input information. In this embodiment, the dialog model is used to analyze the input information to generate at least one response message matching the input information. Therefore, after the question-answering system detects input information in a scenario task, the dialog model matching the scenario task can be retrieved. Because the dialog model's internal state information can help the large model better understand the user's input information, the dialog model's internal state information can be used to analyze the input information, enabling the large model to generate at least one response message matching the input information. In step S303, target text features of the response message are obtained from the dialog model's internal state information. In the technical solution provided in step S303 of the present disclosure, the internal state information of the dialogue model is used to represent the rules by which the dialogue model analyzes input information. Based on this, after analyzing the input information using the internal state information of the dialogue model, the target text features of the reply information can be obtained from the internal state information of the dialogue model. The target text features represent the semantics of the reply information within the scenario task. In this embodiment, the internal state information of the dialogue model may include the rules by which multiple network layers within the dialogue model analyze the input information. The multiple network layers of the dialogue model may include a decoder layer, a fully connected layer (FC layer), etc. These multiple network layers are used to perform different processing operations on the input information to generate output results. The target text features of the reply information can be obtained from the output results of the multiple network layers. The method for obtaining the target text features of the reply information can be referred to the description of step S203 above and will not be repeated here. Step S304: Determine the detection result of the reply information based on the target text features of the reply information. In the technical solution provided in step S304 of the present disclosure, after obtaining the target text features of the reply information, a reply information detection result can be determined based on the target text features of the reply information. The detection result indicates whether the reply information output by the dialogue model is reliable. In this embodiment, after obtaining the target text features of the reply information, a covariance matrix can be constructed based on the target text features of the reply information. A metric for the reply information output by the dialogue model can then be calculated based on the covariance matrix. The metric can be a hallucination degree measurement score corresponding to the reply information output by the dialogue model.After obtaining the metric, the reliability of the response information output by the dialogue model can be further determined based on the metric and the metric threshold. The specific method for determining the reliability of the response information output by the dialogue model based on the metric and the metric threshold can be found in the description of step S204 above and will not be repeated here. In step S305, in response to the detection result indicating that the response information output by the dialogue model is reliable in the scenario task, the response information is output. In the technical solution provided in step S305 above, after determining the response information detection result, whether to output the response information can be determined based on whether the response information output by the dialogue model is reliable in the scenario task as indicated by the detection result. In this embodiment, if the detection result indicates that the response information output by the dialogue model is reliable in the scenario task, it indicates that the response information output by the dialogue model has a high accuracy rate. In this case, the response information can be output. In step S306, in response to the detection result indicating that the response information output by the dialogue model is unreliable in the scenario task, a corresponding prompt message is output. In the technical solution provided in step S306 of the present disclosure, unlike step S305, if the detection result indicates that the response information output by the dialogue model is unreliable for the scenario task, it indicates that the accuracy of the response information output by the dialogue model is low. In this case, the response information may not be output, but a corresponding prompt message may be output to remind the user that the response information output by the dialogue model does not match the input information. Based on steps S301 to S306 of the above embodiment, the internal state information of the dialogue model is utilized to obtain the target text features of the response information output by the dialogue model. This can better mine and utilize the semantic features of the response information, and then use the target text features to determine the response information detection result. This detection result can indicate whether the response information is reliable for the scenario task. If the response information is reliable for the scenario task, the response information is output. If the response information is unreliable for the scenario task, a prompt message is output to ensure that only response information that has been verified to be reliable is output. This achieves the technical effect of effectively detecting the reliability of the response information output by the dialogue model, thereby solving the technical problem of being unable to effectively detect the response information output by the model. According to an embodiment of the present disclosure, a method for detecting information is also provided in terms of human-computer interaction. FIG4 is a flow chart of another method for detecting information according to an embodiment of the present disclosure. As shown in FIG4 , the method may include the following steps: Step S401: Monitoring input information on a dialogue interface.In the technical solution provided in step S401 above, the dialogue interface can be an interface for users to interact with the question-and-answer system. The user can enter the input information on the dialogue interface. The input information can be multimodal information, and the types of multimodal information include at least one of the following: text information containing character information, video frame information containing frame image information, and audio information. The question-and-answer system can monitor the input information on the dialogue interface. In step S402, at least one reply message matching the input information is displayed on the dialogue interface. In the technical solution provided in step S402 above, after the input information is monitored on the dialogue interface, the dialogue model can be retrieved to analyze the input information, thereby obtaining at least one reply message matching the input information, and the at least one reply message matching the input information is displayed on the dialogue interface. The reply information can be multimodal information, and the types of the reply information can include at least one of the following: text information, image information, video information, and voice information. The reply information is obtained by analyzing the input information using the internal state information of the dialogue model. The process of analyzing the input information using the internal state information of the dialogue model can be referenced to the description of step S202 above and will not be repeated here. Step S403: In response to an information detection operation performed on the dialogue interface, the detection result of the reply information is displayed on the dialogue interface. In the technical solution provided in step S403 of the present disclosure, in response to the information detection operation on the dialogue interface, the detection result of the reply information can be displayed on the dialogue interface. This detection result indicates whether the reply information output by the dialogue model is reliable and is determined based on the target text features of the reply information. The target text features represent the semantics of the reply information and are obtained from the internal state information of the dialogue model. This internal state information represents the rules used by the dialogue model to analyze the input information. The method for determining whether the reply information output by the dialogue model is reliable can be referenced to the description of steps S203 and S204 above and will not be repeated here. Based on steps S401 to S403 of the above embodiment, input information is monitored on the dialogue interface; at least one piece of reply information matching the input information is displayed on the dialogue interface; and in response to an information detection operation performed on the dialogue interface, a detection result of the reply information is displayed on the dialogue interface, wherein the detection result is used to indicate whether the reply information output by the dialogue model is reliable.That is, the response information detection results can be intuitively displayed through the dialogue interface, allowing users to intuitively understand whether the response information output by the dialogue model is reliable. Users can quickly obtain the required information, improving the user experience. In addition, the display of the detection results can help users determine the reliability of the response information, helping to improve the accuracy and credibility of the human-computer dialogue. Figure 5 is a schematic diagram of an information detection system according to an embodiment of the present disclosure. As shown in Figure 5, the information detection system 500 includes: an information input terminal 501, an information detection terminal 502, and an information output terminal 503. The information input terminal 501 is configured to monitor input information. In this embodiment, the user can input information into the information input terminal, which in turn monitors the user's input information. The input information can be multimodal information, for example, text information containing characters, video frame information containing frame image information, audio information, etc., without specific limitation herein. The information detection terminal 502 is configured to input input information into the dialogue model for analysis, obtain at least one reply matching the input information, and obtain target text features of the reply from the dialogue model's internal state information. The internal state information represents the dialogue model's analysis rules for the input information, and the target text features represent the semantics of the reply. Based on the target text features of the reply, a detection result is determined for the reply. The detection result indicates whether the reply output by the dialogue model is reliable. In this embodiment, after the information input terminal detects the input information, it can transmit the input information to the information detection terminal. Upon receiving the input information, the information detection terminal can retrieve the dialogue model and analyze the input information using the dialogue model's internal state information to obtain at least one reply matching the input information. Based on the target text features of the reply, a detection result is determined for the reply, and the reliability of the reply output by the dialogue model is determined based on the detection result. The step of analyzing the input information using the internal state information of the dialogue model to obtain at least one piece of reply information matching the input information can be found in the description of step S202 above and will not be repeated here. The step of obtaining the target text features of the reply information from the internal state information of the dialogue model can be found in the description of step S203 above and will not be repeated here. The step of determining the detection result of the reply information based on the target text features of the reply information can be found in the description of step S204 above and will not be repeated here. The information output terminal 503 is configured to output the reply information in response to a detection result indicating that the reply information output by the dialogue model is reliable; and to output a corresponding prompt message in response to a detection result indicating that the reply information output by the dialogue model is unreliable.In this embodiment, after determining the detection result of the reply information, the information detection terminal may transmit the detection result to the information output terminal. After receiving the detection result, the information output terminal may determine whether the reply information output by the dialogue model is reliable based on the detection result. If the detection result indicates that the reply information output by the dialogue model is reliable, the information output terminal outputs the reply information. Conversely, if the detection result indicates that the reply information output by the dialogue model is unreliable, the information output terminal may output a prompt message to inform the user that the reply information output by the dialogue model is unreliable. The prompt message may be a text prompt message, for example, "The reply information output by the dialogue model is unreliable," or an image message. The specific content of the prompt message is not limited herein. The following further describes the technical solutions of the embodiments of this disclosure with examples in conjunction with preferred embodiments. Currently, the requirements for the authenticity and reliability of reply information output by large models are constantly increasing. When a large model outputs content that is "inconsistent with the facts" or "fabricated out of thin air," it is considered that the large model output contains knowledge illusions. Typical large models inevitably output erroneous and unfounded illusions. This potential uncertainty and unreliability poses serious challenges to the commercial implementation and application of large models. Therefore, it is particularly important to determine whether the output of the large model is reliable and accurate. In one embodiment, a preplexity-based hallucination detection method or a self-check (SelfCheckGPT)-based hallucination detection method can be used to determine the reliability of the response information output by the large model. The preplexity-based hallucination detection method primarily determines the uncertainty of each character in the response information output by the large model, and then multiplies the uncertainty of each character to obtain the uncertainty of the response information output by the large model. However, due to inaccurate uncertainty estimation of the sentence length of the response information output by the large model and the diversity of sentence expression forms in the response information output by the large model, the reliability detection results of the response information output by the large model may be inaccurate. The hallucination detection method based on SelfCheckGPT determines the reliability of the response information output by the large model by measuring the consistency of the response information output by the large model multiple times. However, this method requires an additional large model to calculate the consistency of the response information output multiple times, resulting in a large computational time overhead. In addition, this method can only detect self-contradictory hallucinations output by the large model, and cannot detect hallucinations with high consistency output by the large model. Therefore, the above methods all have the technical problem of being unable to effectively detect the response information output by the large model.However, an embodiment of the present disclosure provides a method for detecting information, which obtains input information input by a user and allows a large model to output multiple reply information based on the input information input by the user; obtains a text feature set of the reply information of the penultimate layer of the internal output layer of the large model; uses a dynamic feature clipping scheme to clip text features of the text feature set that exceed a threshold to remove abnormal feature responses; obtains text features of the last text unit of the intermediate layer of different output hidden states of the large model; constructs a feature covariance matrix for the text features of the last text unit of different output layers of the large model; calculates a metric based on the covariance matrix; and determines whether the reply information output by the large model has hallucinations based on the metric, wherein when the metric is greater than the metric threshold, it is determined that the reply information output by the large model has knowledge hallucinations, that is, the reply information output by the large model is unreliable, and when the metric is not greater than the metric threshold, it is determined that the reply information output by the large model does not have knowledge hallucinations, that is, the reply information output by the large model is reliable. In other words, in this embodiment of the present disclosure, a metric can be directly used to measure the inconsistency of multiple response messages output by the large model. This is equivalent to measuring the continuous entropy of semantics in the feature space. Using this metric can better characterize the uncertainty of the response messages output by the large model, thereby reflecting the degree of knowledge illusion in the response messages. In other words, it reflects whether the response messages output by the large model are reliable. This achieves the technical effect of effectively testing the reliability of the response messages output by the large model, thereby resolving the technical problem of being unable to effectively test the reliability of the response messages output by the large model. The information detection method in this embodiment of the present disclosure is now further described. Figure 6 is a flowchart of an information detection method according to an embodiment of the present disclosure. As shown in Figure 6, the method may include the following steps: Step S601: Obtaining a user question. In this embodiment, a user can enter any question on the operation interface, and the large model can monitor the user question entered on the operation interface to obtain the user question. Step S602: Outputting 10 response messages based on the user question. In this embodiment, after receiving the user's input question, the large model can output 10 responses based on the user's question. These 10 responses are information that answers the user's question. Step S603 obtains the text feature set output by the penultimate layer of the large model's internal output layer. In this embodiment, the large model is composed of two parts: the decoder layer and the final fully connected layer (FC layer). The penultimate layer serves as the output of the decoder layer and the input of the fully connected layer (FC layer).The output of the decoder layer is the text feature representation of the text unit token in the reply message. The text features of the text unit token in the reply message can be obtained to form a text feature set. In step S604, a dynamic feature clipping scheme is used to clip portions of the text feature set that exceed a threshold. In this embodiment, the dynamic clipping scheme can be used to clip abnormal feature responses in the text feature set that exceed the threshold, thereby removing abnormal feature responses from the text feature set. Figure 7 is a schematic diagram of feature responses in a text feature set according to an embodiment of the present disclosure. As shown in Figure 7, the feature distribution of the text feature set at the penultimate layer of a large model is shown. As shown in Figure 7, the text feature set contains a large number of extreme feature responses. These extreme feature responses can easily cause the large model to output feature classification results with excessively high confidence. Therefore, to mitigate the highly consistent hallucinatory outputs caused by oversaturated classification, feature clipping can be used to clip abnormal feature responses from the text feature set. Optionally, feature clipping can be performed using the following piecewise function: when the feature response h exceeds the maximum threshold hmax, the response is truncated to the maximum threshold hmax; when the feature response h is less than the minimum threshold hmin, the response is truncated to the minimum threshold hmin. The piecewise function is shown in the aforementioned formula (1) and will not be described in detail here. Optionally, the minimum threshold hmin and the maximum threshold hmax can be obtained by dynamically storing feature responses in a MemoryBank. The MemoryBank is a first-in-first-out feature memory that can store feature responses of 3,000 text units. The feature responses of the 3,000 text units are sorted, and hmax is set to the 0.2% percentile of the maximum response amplitude, and hmin is set to the 0.2% percentile of the minimum response amplitude. This is only an example and does not limit the method for determining the minimum threshold hmin and the maximum threshold hmax. Step S605: Obtain the text features of the last text unit in the intermediate layer of different output hidden states of the large model. In this embodiment, for each user question, the model can output 10 answers (10 sentence outputs). Each answer contains multiple text units, that is, each answer contains multiple words (tokens). Each word (token) in the 10 sentences corresponds to a feature vector in different layers of the large model.Since in large models, the features of the last word (token) of a sentence often contain the information of the entire sentence, therefore, the features of the last word (token) are commonly used as the features of the entire sentence. Based on this, the features of the last word (token) of the intermediate layer can be used as the features of the entire sentence. For example, the llama7B model has 33 layers, and the features of the last word (token) of the 17th layer can be selected as the features of the entire sentence. That is, for the output of 10 sentences, the intermediate layer features of the last word (token) of each sentence are obtained as the features of these 10 sentences respectively. Step S606, construct a sentence feature covariance matrix for the text features of the last text unit of different outputs of the large model. In this embodiment, after obtaining the text features of the last text unit of different outputs of the large model, the sentence feature covariance matrix can be constructed using these text features. The feature representation of the i-th answer of the model can be Zj = h. T o Then for the K generated answers, the covariance matrix of the sentences can be represented by the following formula (5):

[0006] £ = 2匚启 (5) where Z is used to represent the sentence feature matrix, ZT can be used to represent the transpose matrix of the sentence feature matrix, and Jd can be a column vector of all 1s. Optionally, the sentence feature matrix Z can be obtained by concatenating the text features of the last text unit of different answers of the large model. For example, Z = [Z1, Z2,... Zj..., Z k , where Zi can be used to represent the text features corresponding to the i-th answer of the large model. Step S607, calculate the hallucination degree metric according to the covariance matrix. In this embodiment, after determining the covariance matrix, the hallucination degree metric can also be calculated according to the covariance matrix, where the hallucination degree metric can be the hallucination degree metric score. For example, after obtaining the covariance matrix, the hallucination degree metric can be calculated by the following formula (6) to obtain the logarithm determinant of the covariance matrix.

[0007] E = -logdet(S + aI K ) (6)

[0008] K Where, E can be used to represent the hallucination degree metric, K can be used to represent the number of answers output by the large model, £ can be used to represent the covariance matrix, OCIK is a small regularization term used to prevent the covariance matrix from being a non-full rank matrix, Qing can be used to represent the identity matrix with latitude K, and a is a characteristic parameter that can be 0.00L. Optionally, since the determinant of the matrix can be obtained by calculating the eigenvalue, the metric in the above formula can be expressed as:

[0009] E = Hog(rii) = jlX"og(i) (7) where, i = {i], i2, i can be used to represent the matrix S + a I KK eigenvalues ​​of the K features. Step S608: Determine whether the output of the large model contains knowledge hallucinations based on a hallucination degree measurement index. In this step, after obtaining the hallucination degree measurement index, the hallucination degree measurement index can be compared with an index threshold. If the hallucination degree measurement index is greater than the index threshold, it indicates that the output of the large model contains knowledge hallucinations. If the hallucination degree measurement index is not greater than the index threshold, it indicates that the output of the large model does not contain knowledge hallucinations. In the above steps S601 to S608, the questions input by the user are obtained, and the large model is allowed to output multiple answer information according to the questions input by the user; the text feature set of the answer information of the second-to-last layer of the internal output layer of the large model is obtained; the text features of the text feature set that exceed the threshold are clipped using a dynamic feature clipping scheme to remove abnormal feature responses; the text features of the last text unit of the intermediate layer of different output hidden states of the large model are obtained; a feature covariance matrix is ​​constructed for the text features of the last text unit of different output layers of the large model; a metric is calculated based on the covariance matrix; and the metric is used to determine whether the answer information output by the large model has hallucinations, wherein when the metric is greater than the metric threshold, it is determined that the answer information output by the large model has knowledge hallucinations, that is, the answer information output by the large model is unreliable, and when the metric is not greater than the metric threshold, it is determined that the answer information output by the large model does not have knowledge hallucinations, that is, the answer information output by the large model is reliable. In other words, in this embodiment of the present disclosure, a metric can be directly used to measure the inconsistency of multiple responses output by the large model. This is equivalent to measuring the continuous entropy of semantics in the feature space. Using this metric can better characterize the uncertainty of the responses output by the large model, thereby reflecting the degree of knowledge hallucination in the responses. In other words, it reflects whether the responses output by the large model are reliable. This effectively verifies the reliability of responses output by the large model, thereby resolving the technical problem of being unable to effectively verify the reliability of responses output by the large model. Figure 8 is a schematic diagram of large-model knowledge hallucination detection according to an embodiment of the present disclosure. As shown in Figure 8, when a user inputs the question "What was the specific date when XXX first landed on the moon in 1969?", after receiving the user's input, the large language model (LLM model) can process the question using multiple internal network layers and output K different responses. For example, answer 1: hallucination degree metric score a; answer 2: hallucination degree metric score b; answer K: hallucination degree metric score c. By analyzing the K different answers, K hallucination degree measurement scores are obtained, wherein the K hallucination degree measurement scores are used to measure the reliability of the K answers respectively.When the hallucination degree metric score of an answer exceeds a threshold, the answer corresponding to the hallucination degree metric score is output. If no answer has a hallucination degree metric score exceeding the threshold, only a prompt message is output, for example, indicating that the answer to the question is not supported. It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, and displayed data, etc.) involved in this disclosure are all authorized by the user or fully authorized by all parties. The collection, use, and processing of relevant data must comply with relevant laws, regulations, and standards of relevant countries and regions, and corresponding operation portals are provided for users to choose to authorize or deny. It should be noted that for the sake of simplicity, the aforementioned method embodiments are described as a series of combined actions. However, those skilled in the art should be aware that this disclosure is not limited by the order of the actions described, as certain steps can be performed in other orders or simultaneously according to this disclosure. Secondly, those skilled in the art should also be aware that the embodiments described in this specification are all preferred embodiments, and the actions and components involved are not necessarily required for this disclosure. Through the above description of the embodiments, those skilled in the art will clearly understand that the methods according to the above embodiments can be implemented using software and a necessary general-purpose hardware platform, or alternatively, hardware. Based on this understanding, the technical solution of the present disclosure, or the portion that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, a magnetic disk, or an optical disk) and includes instructions for enabling a terminal device (which may be a mobile phone, a computer, a server, or a network device, etc.) to execute the methods described in the various embodiments of the present disclosure. According to an embodiment of the present disclosure, an information detection device for implementing the above-mentioned information detection method is also provided. FIG9 is a schematic diagram of an information detection device according to an embodiment of the present disclosure. As shown in FIG9 , the information detection device 900 includes a first monitoring component 901, a first analysis component 902, a first acquisition component 903, and a first determination component 904. The first monitoring component 901 is configured to monitor input information. The first analysis component 902 is configured to input the input information into a dialogue model for analysis to obtain at least one piece of response information that matches the input information. The first acquisition component 903 is configured to acquire target text features of the reply information from the internal state information of the dialogue model, wherein the internal state information is used to represent the rules for the dialogue model to analyze the input information, and the target text features are used to represent the semantics of the reply information.The first determination component 904 is configured to determine a detection result of the reply information based on target text features of the reply information, where the detection result is used to indicate whether the reply information output by the dialogue model is reliable. It should be noted that the first monitoring component 901, the first analysis component 902, the first acquisition component 903, and the first determination component 904 correspond to steps S201 to S204 in the aforementioned embodiment. The examples and application scenarios implemented by these four components and the corresponding steps are the same, but are not limited to the content disclosed in the first embodiment. It should be noted that the above-mentioned components or assemblies can be hardware components or software components stored in a memory (e.g., memory 104) and processed by one or more processors (e.g., processors 102a, 102b, 102n). The above-mentioned components can also be executed as part of the device in the computer terminal 10 provided in the first embodiment. According to an embodiment of the present disclosure, an information generation device for implementing the above-mentioned information generation method is also provided. FIG10 is a schematic diagram of another information generation device according to an embodiment of the present disclosure. As shown in FIG10 , the information generation device 1000 may include: a second monitoring component 1001, a second analysis component 1002, a second acquisition component 1003, a second determination component 1004, a first output component 1005, and a second output component 1006. The second monitoring component 1001 is configured to monitor input information in a scenario task on the user interface of the question-answering system. The second analysis component 1002 is configured to retrieve a dialogue model matching the scenario task, input the input information into the dialogue model for analysis, and obtain at least one reply matching the input information. The second acquisition component 1003 is configured to obtain target text features of the reply information from the internal state information of the dialogue model. The internal state information represents the rules used by the dialogue model to analyze the input information, and the target text features represent the semantic meaning of the reply information in the scenario task. The second determination component 1004 is configured to determine a detection result of the reply information based on target text features of the reply information. The first output component 1005 is configured to output the reply information in response to a detection result indicating that the reply information output by the dialogue model is reliable in the scenario task. The second output component 1006 is configured to output corresponding prompt information in response to a detection result indicating that the reply information output by the dialogue model is unreliable in the scenario task.It should be noted that the aforementioned second monitoring component 1001, second analysis component 1002, second acquisition component 1003, second determination component 1004, first output component 1005, and second output component 1006 correspond to steps S301 to S306 in the aforementioned embodiment. The six components implement the same examples and application scenarios as the corresponding steps, but are not limited to the content disclosed in the aforementioned embodiment 1. It should be noted that the aforementioned components or assemblies may be hardware components or software components stored in a memory (e.g., memory 104) and processed by one or more processors (e.g., processors 102a, 102b, 102n). The aforementioned components may also be part of a device that can run on the computer terminal 10 provided in the first embodiment. According to an embodiment of the present disclosure, an information detection device for implementing the aforementioned information detection method is also provided. FIG11 is a schematic diagram of another information detection device according to an embodiment of the present disclosure. As shown in FIG11, the information detection device 1100 may include a third monitoring component 1101, a first display component 1102, and a second display component 1103. The third monitoring component 1101 is configured to monitor input information on the dialogue interface. The first display component 1102 is configured to display at least one reply message matching the input information on the dialogue interface. The reply message is obtained by analyzing the input information using the internal state information of the dialogue model. The second display component 1103 is configured to respond to an information detection operation performed on the dialogue interface and display the reply message detection result on the dialogue interface. The detection result indicates whether the reply message output by the dialogue model is reliable and is determined based on the target text features of the reply message. The target text features represent the semantics of the reply message and are obtained from the internal state information of the dialogue model. The internal state information represents the rules used by the dialogue model to analyze the input information. It should be noted that the third monitoring component 1101, the first display component 1102, and the second display component 1103 correspond to steps S401 to S403 in the aforementioned embodiment. The examples and application scenarios implemented by these three components and the corresponding steps are the same, but are not limited to the content disclosed in the aforementioned embodiment 1. It should be noted that the above-mentioned components or components may be hardware components or software components stored in a memory (e.g., memory 104) and processed by one or more processors (e.g., processors 102a, 102b, 102n). The above-mentioned components may also be run as part of a device in the computer terminal 10 provided in the first embodiment.It should be noted that the preferred implementation schemes involved in the above-mentioned embodiments of the present disclosure are the same as the schemes, application scenarios, and implementation processes provided in the above-mentioned embodiments, but are not limited to the schemes provided in the above-mentioned embodiments. The embodiments of the present disclosure may provide a computer terminal, which may be any computer terminal device in a computer terminal group. Optionally, in this embodiment, the computer terminal may be replaced by a terminal device such as a mobile terminal. Optionally, in this embodiment, the computer terminal may be located on at least one of multiple network devices in a computer network. In this embodiment, the computer terminal may execute the program code for the following steps in the information detection method: monitoring input information; inputting the input information into a dialogue model for analysis to obtain at least one reply message matching the input information; obtaining target text features of the reply message from the internal state information of the dialogue model, wherein the internal state information represents the rules used by the dialogue model to analyze the input information, and the target text features represent the semantics of the reply message; and determining a reply message detection result based on the target text features of the reply message, wherein the detection result indicates whether the reply message output by the dialogue model is reliable. Optionally, Figure 12 is a block diagram of the structure of a computer terminal according to an embodiment of the present disclosure. As shown in Figure 12 , the computer terminal A may include one or more processors 1202 (only one is shown), a memory 1204, a storage controller, and a peripheral interface. The peripheral interface is connected to a radio frequency component, an audio component, and a display. The memory may be used to store software programs and components, such as the program instructions / components corresponding to the reply information detection method and apparatus in the embodiments of the present disclosure. The processor executes the software programs and components stored in the memory to perform various functional applications and data processing, thereby implementing the reply information detection method described above. The memory may include high-speed random access memory (RAM) and non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory may further include memory located remotely from the processor, which may be connected to the terminal A via a network. Examples of such networks include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.The processor can access information and applications stored in the memory through a transmission device to perform the following steps: monitoring input information; inputting the input information into a dialogue model for analysis to obtain at least one reply message matching the input information; obtaining target text features of the reply message from internal state information of the dialogue model, wherein the internal state information represents the rules used by the dialogue model to analyze the input information, and the target text features represent the semantics of the reply message; and determining a detection result of the reply message based on the target text features of the reply message, wherein the detection result indicates whether the reply message output by the dialogue model is reliable. Optionally, the processor can also execute program code for the following steps: obtaining a text feature set of the reply message in the network layer of the dialogue model, wherein the internal state information includes a text feature set in the network layer, and the text feature set in the network layer includes text features of text units constituting the reply message; and determining the target text feature from the text feature set in the network layer. Optionally, the processor may further execute program code for the following steps: determining, from a text feature set in a network layer, text features corresponding to a target text unit in the reply message, wherein the target text unit includes the semantics of the reply message; and determining the text features corresponding to the target text unit as target text features. Optionally, the processor may further execute program code for the following steps: obtaining a text feature set of the reply message in an output network layer of a dialogue model; determining, in a target hidden network layer of the dialogue model, a text feature set of a target hidden network layer corresponding to the text feature set of the output network layer. Optionally, the processor may further execute program code for the following steps: updating the text feature set of the output network layer to obtain an updated text feature set, wherein the updated text feature set does not include noise features; and determining, in the target hidden network layer, a text feature set of the target hidden network layer corresponding to the updated text feature set. Optionally, the processor may further execute program code for the following steps: in response to a response amplitude of a text feature in the text feature set of the output network layer being less than a first response amplitude threshold, adjusting the response amplitude to the first response amplitude threshold; in response to a response amplitude of a text feature in the text feature set of the output network layer being greater than or equal to the first response amplitude threshold and less than or equal to a second response amplitude threshold, maintaining the response amplitude, wherein the second response amplitude threshold is greater than or equal to the first response amplitude threshold; in response to a response amplitude of a text feature in the text feature set of the output network layer being greater than a second response amplitude threshold, adjusting the response amplitude to the second response amplitude threshold; and determining the response amplitude adjusted to the first response amplitude threshold, the maintained response amplitude, and the response amplitude adjusted to the second response amplitude threshold as an updated text feature set.Optionally, the processor may further execute program code for the following steps: obtaining the order of arrangement of the multiple hidden network layers in the dialogue model; and determining a target hidden network layer from the multiple hidden network layers based on the order of arrangement. Optionally, the processor may further execute program code for the following steps: determining a hidden network layer with a middle order of arrangement as the target hidden network layer. Optionally, the processor may further execute program code for the following steps: determining a metric for the reply information based on target text features of the reply information, wherein the metric is used to indicate the reliability of the reply information output by the dialogue model; and determining a detection result for the reply information based on the metric. Optionally, the processor may further execute program code for the following steps: in response to the metric being greater than a metric threshold, determining a detection result that the reply information output by the dialogue model is unreliable; and in response to the metric being less than or equal to the metric threshold, determining a detection result that the reply information output by the dialogue model is reliable. Optionally, the processor may further execute program code for the following steps: if the detection result indicates that the reply information output by the dialogue model is unreliable, prohibiting the output of the reply information; and if the detection result indicates that the reply information output by the dialogue model is reliable, outputting the reply information. Optionally, the processor may further execute program code for the following steps: constructing a feature covariance matrix using multiple target text features of multiple reply information; and determining a metric for the reply information based on the feature covariance matrix. Embodiments of the present disclosure provide an information detection method. By utilizing internal state information of the dialogue model to obtain target text features of the reply information output by the dialogue model, semantic features of the reply information can be better mined and utilized. Furthermore, the target text features can be used to determine a reply information detection result. This detection result can represent the overall uncertainty of the reply information, that is, reflect whether the reply information output by the dialogue model is reliable. This achieves the technical effect of effectively detecting the reliability of the reply information output by the dialogue model, thereby resolving the technical problem of being unable to effectively detect the reliability of reply information output by a large model. Those skilled in the art will appreciate that the structure shown in FIG12 is merely illustrative, and that the computer terminal may also be a smartphone (e.g., an Android phone, an iOS phone, etc.), a tablet computer, a PDA, a mobile internet device (MID), a PAD, or other terminal device. FIG12 does not limit the structure of these electronic devices. For example, computer terminal A may include more or fewer components (e.g., a network interface, a display device, etc.) than those shown in FIG12 , or may have a configuration different from that shown in FIG12 .Those skilled in the art will appreciate that all or part of the steps in the various methods of the above embodiments can be completed by instructing the hardware of the terminal device through a program. The program can be stored in a computer-readable storage medium, which may include a flash drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk. The embodiments of the present disclosure also provide a computer-readable storage medium. Optionally, in this embodiment, the storage medium can be used to store the program code executed by the reply information detection method provided in the first embodiment. Optionally, in this embodiment, the storage medium can be located in any computer terminal in a computer terminal group in a computer network, or in any mobile terminal in a mobile terminal group. Optionally, in this embodiment, the storage medium is configured to store program code for performing the following steps: monitoring input information; inputting the input information into a dialogue model for analysis to obtain at least one reply information matching the data information; obtaining target text features of the reply information from the internal state information of the dialogue model, wherein the internal state information represents the rules used by the dialogue model to analyze the input information, and the target text features represent the semantics of the reply information; and determining a detection result of the reply information based on the target text features of the reply information, wherein the detection result indicates whether the reply information output by the dialogue model is reliable. Embodiments of the present disclosure also provide a computer program product comprising computer instructions that, when executed by a processor, implement the information detection method provided in embodiments of the present disclosure. In this embodiment, the computer instructions may be stored in a read-only memory (ROM) or loaded from a storage component into a random access memory (RAM) to enable the processor to perform various appropriate actions and processes in the database product status detection method. In some embodiments, some or all of the computer instructions may be loaded and / or installed on an electronic device via a read-only memory and / or a communication component. When computer instructions are loaded into random access memory and executed by a computing component, one or more steps of the information detection method described above can be performed. The serial numbers of the embodiments of the present disclosure are for illustrative purposes only and do not represent the superiority or inferiority of each embodiment. In the above embodiments of the present disclosure, the descriptions of each embodiment are given with emphasis. For details not provided in one embodiment, please refer to the relevant descriptions of other embodiments. In the several embodiments provided in this disclosure, it should be understood that the disclosed technical content can be implemented in other ways.The device embodiments described above are merely illustrative. For example, the component divisions described represent only one logical functional division. In actual implementation, different divisions may be employed. For example, multiple components or components may be combined or integrated into another system, or some features may be omitted or not implemented. Furthermore, the couplings or direct couplings or communication connections shown or discussed may be through interfaces, or indirect couplings or communication connections between components may be electrical or other. Components described as separate components may or may not be physically separate, and components shown as components may or may not be physical components, i.e., they may be located in one location or distributed across multiple network components. Some or all of these components may be selected to achieve the objectives of the present embodiments as needed. Furthermore, the functional components in the various embodiments of the present disclosure may be integrated into a single processing component, each component may exist physically separately, or two or more components may be integrated into a single component. These integrated components may be implemented in either hardware or software functional components. If the integrated components are implemented as software functional components and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this disclosure, or the portion that contributes to the prior art, or all or part of this technical solution, can be embodied in the form of a software product. This computer software product, stored in a storage medium, includes instructions for enabling a computer device (such as a personal computer, server, or network device) to perform all or part of the steps of the methods described in various embodiments of this disclosure. The aforementioned storage media include various media capable of storing program code, such as USB flash drives, read-only memories (ROMs), random access memories (RAMs), removable hard drives, magnetic disks, or optical disks. The above description is merely a preferred embodiment of this disclosure. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of this disclosure, and such improvements and modifications should also be considered within the scope of protection of this disclosure.Industrial Applicability The solution provided by the embodiment of the present disclosure can be applied to a question-answering system. By analyzing the input information based on the dialogue model, at least one reply information matching the input information can be obtained. Since the internal state information of the dialogue model is used to represent the rules for the dialogue model to analyze the input information, the target text features of the reply information can be obtained from the internal state information of the dialogue model, thereby better mining and utilizing the semantic features of the reply information. The target text features are used to determine the detection result of the reply information. The detection result can characterize the overall uncertainty of the reply information, that is, reflect whether the reply information output by the dialogue model is reliable, thereby achieving the technical effect of effectively performing reliability detection on the reply information output by the dialogue model, thereby solving the technical problem of being unable to effectively perform reliability detection on the reply information output by a large model.

Claims

Claims 1. A method for detecting information, comprising: Monitor the input information; Input the input information into a dialogue model for analysis to obtain at least one reply information that matches the input information; obtain the target text feature of the reply information from the internal state information of the dialogue model, where the internal state information is used to represent the rule for the dialogue model to analyze the input information, and the target text feature is used to represent the semantics of the reply information; determine the detection result of the reply information based on the target text feature of the reply information, where the detection result is used to represent whether the reply information output by the dialogue model is reliable.

2. The method according to claim 1, wherein Obtain the target text feature corresponding to the reply information from the internal state information of the dialogue model, including: obtaining the text feature set of the reply information in the network layer of the dialogue model, where the internal state information includes the text feature set in the network layer, and the text feature set in the network layer includes the text features of the text units constituting the reply information; determine the target text feature from the text feature set in the network layer.

3. The method according to claim 2, wherein, Determine the target text feature from the text feature set in the network layer, including: determining the text feature corresponding to the target text unit in the reply information from the text feature set in the network layer, where the target text unit includes the semantics of the reply information; determine the text feature corresponding to the target text unit as the target text feature.

4. The method according to claim 2, wherein Obtain the text feature set of the reply information in the network layer of the dialogue model, including: obtaining the text feature set of the reply information in the output network layer of the dialogue model; determining the text feature set of the target hidden network layer corresponding to the text feature set of the output network layer in the target hidden network layer of the dialogue model.

5. The method according to claim 4, wherein Determine the text feature set of the target hidden network layer corresponding to the text feature set of the output network layer in the target hidden network layer of the dialogue model, including: updating the text feature set of the output network layer to obtain the updated text feature set, where the updated text feature set does not include noise features; determining the text feature set of the target hidden network layer corresponding to the updated text feature set in the target hidden network layer.

6. The method according to claim 5, wherein Update the text feature set of the output network layer 29 Obtain the updated text feature set, including: in response to the response amplitude of the text features in the text feature set of the output network layer being less than the first response amplitude threshold, adjust the response amplitude to the first response amplitude threshold; in response to the response amplitude of the text features in the text feature set of the output network layer being greater than or equal to the first response amplitude threshold and less than or equal to the second response amplitude threshold, keep the response amplitude, where the second response amplitude threshold is greater than or equal to the first response amplitude threshold; in response to the response amplitude of the text features in the text feature set of the output network layer being greater than the second response amplitude threshold, adjust the response amplitude to the second response amplitude threshold; determine the adjusted first response amplitude threshold, the kept response amplitude, and the adjusted second response amplitude threshold as the updated text feature set.

7. The method according to claim 4, wherein The method further includes: respectively obtaining the arrangement order of multiple hidden network layers in the dialogue model; based on the arrangement order, determining the target hidden network layer in the multiple hidden network layers.

8. The method according to claim 7, wherein Based on the arrangement order, determining the target hidden network layer in the multiple hidden network layers includes: determining the hidden network layer in the middle arrangement order as the target hidden network layer.

9. The method according to claim 1, wherein Based on the target text features of the reply information, determining the detection result of the reply information includes: based on the target text features of the reply information, determining a metric for the reply information, where the metric is used to represent the reliability of the dialogue model outputting the reply information; determining the detection result of the reply information based on the metric.

10. The method according to claim 9, wherein Based on the metric, determining the detection result of the reply information includes: in response to the metric being greater than the metric threshold, determining that the detection result is that the reply information output by the dialogue model is unreliable; in response to the metric being less than or equal to the metric threshold, determining that the detection result is that the reply information output by the dialogue model is reliable.

11. The method according to claim 10, wherein The method further includes: in the case where the detection result is that the reply information output by the dialogue model is unreliable, prohibiting the output of the reply information; in the case where the detection result is that the reply information output by the dialogue model is reliable, outputting the reply information.

12. The method according to claim 9, wherein Based on the target text features of the reply information, determining the metric for the reply information includes: 30 Using at least one target text feature of at least one reply information to construct a feature covariance matrix; based on the feature covariance matrix, determining the metric for the reply information.

13. A method for generating information, which is applied to a question-and-answer system deployed in a scenario task, and the method includes: On the operation interface of the question-and-answer system, monitor the input information in the scenario task; Retrieve a dialogue model matching the scenario task, input the input information into the dialogue model for analysis, and obtain at least one reply information matching the input information; obtain the target text feature of the reply information from the internal state information of the dialogue model, where the internal state information is used to represent the rules for the dialogue model to analyze the input information, and the target text feature is used to represent the semantics of the reply information in the scenario task; determine the detection result of the reply information based on the target text feature of the reply information; in response to the detection result indicating that the reply information output by the dialogue model is reliable in the scenario task, output the reply information; in response to the detection result indicating that the reply information output by the dialogue model is unreliable in the scenario task, output the corresponding prompt information.

14. A method for detecting information, comprising: Monitor the input information on the dialogue interface; On the dialogue interface, display at least one reply information matching the input information, where the reply information is obtained by analyzing the input information using a dialogue model; in response to an information detection operation on the dialogue interface, display the detection result of the reply information on the dialogue interface, where the detection result is used to indicate whether the reply information output by the dialogue model is reliable and is determined based on the target text feature of the reply information, the target text feature is used to represent the semantics of the reply information, and is obtained from the internal state information of the dialogue model, and the internal state information is used to represent the rules for the dialogue model to analyze the input information.

15. The method according to claim 14, wherein, The input information and the reply information are multimodal information, and the types of the multimodal information include at least one of the following: text information containing character information, video frame information containing frame image information, audio information, and the types of the reply information include at least one of the following: text information, image information, video information, and voice information.

16. An information detection system, comprising: An information input end, used to monitor the input information; An information detection end, used to input the input information into a dialogue model for analysis to obtain at least one reply information matching the input information; Obtain the target text feature of the reply information from the internal state information of the dialogue model, where the internal state information is used to represent the rules for the dialogue model to analyze the input information, and the target text feature is used to represent the semantics of the reply information; determine the detection result of the reply information based on the target text feature of the reply information, where The detection result is used to indicate whether the reply information output by the dialogue model is reliable; an information output end, used to output the reply information in response to the detection result indicating that the reply information output by the dialogue model is reliable; and output the corresponding prompt information in response to the detection result indicating that the reply information output by the dialogue model is unreliable.

17. An electronic device, comprising: Memory and processor; The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions. When the computer-executable instructions are executed by the processor, the steps of the method described in any one of claims 1 to 15 are implemented.

18. A computer-readable storage medium, the computer-readable storage medium includes a stored executable program, wherein when the executable program runs, it controls the device where the computer-readable storage medium is located to execute the method described in any one of claims 1 to 15.

19. A computer program product, including a computer program, the computer program implements the method described in any one of claims 1 to 15 when executed by a processor.

20. A computer program product, including a non-volatile computer-readable storage medium, the non-volatile computer-readable storage medium stores a computer program, and the computer program implements the method described in any one of claims 1 to 15 when executed by a processor.

Images