Text confidence detection method and device, storage medium and electronic equipment

Abstract

The invention discloses a text confidence detection method and device, a storage medium and electronic equipment. The method comprises the following steps: acquiring a to-be-detected domain text; inputting the field text into a preset information generation model, determining semantic information corresponding to each statement contained in the field text through the information generation model, and generating a plurality of atomic facts according to the semantic information; according to the feature code corresponding to each atomic fact, determining a target criterion matched with the atomic fact in a domain knowledge base of a domain to which the domain text belongs; inputting each atomic fact and each target criterion into a preset confidence degree detection model so as to determine the confidence degree corresponding to each atomic fact through the confidence degree detection model; and according to the confidence corresponding to each atomic fact, determining the comprehensive confidence of the domain text. According to the scheme, the accuracy and reliability of the text confidence detection result are effectively improved.

Description

Technical Field

[0001] This specification relates to the field of computer technology, and in particular to a text confidence detection method, device, storage medium, and electronic device. Background Art

[0002] Amid the explosive growth of digital information, text detection, a key technology at the intersection of computer vision and natural language processing, has become a crucial foundation for intelligent systems to perceive and understand visual information. Within the text detection process, quantitatively assessing the authenticity or trustworthiness of text is a key step in ensuring its usability. Accurately measuring text confidence provides a reliable basis for subsequent decision-making.

[0003] However, traditional methods rely heavily on manually annotated data to build confidence assessment models. However, the manual annotation process is subject to high subjectivity, high cost, and difficulty in covering complex scenarios. In addition, existing confidence assessment algorithms are mostly based on a single indicator and lack comprehensive consideration of text semantics and contextual relationships, resulting in low accuracy and reliability of the detected confidence.

[0004] Therefore, how to accurately detect the confidence of text is an urgent problem to be solved. Summary of the Invention

[0005] This specification provides a text confidence detection method, device, storage medium and electronic device to partially solve the above-mentioned problems existing in the prior art.

[0006] This manual adopts the following technical solutions: A text confidence detection method, the method comprising: Get the domain text to be detected; Inputting the domain text into a preset information generation model to determine semantic information corresponding to each sentence contained in the domain text through the information generation model, and generating multiple atomic facts based on the semantic information, wherein each atomic fact is a complete declarative sentence describing an independent event that cannot be separated; According to the feature code corresponding to each atomic fact, a target criterion matching the atomic fact is determined in the domain knowledge base of the domain to which the domain text belongs; Inputting each atomic fact and each target criterion into a preset confidence detection model to determine the confidence corresponding to each atomic fact through the confidence detection model; According to the confidence corresponding to each atomic fact, the comprehensive confidence of the domain text is determined.

[0007] Optionally, inputting the domain text into a pre-trained information generation model specifically includes: Based on a preset length, the domain text is split into multiple independent text segments; Each text segment is input into the information generation model.

[0008] Optionally, the information generation model is used to determine semantic information corresponding to each sentence contained in the domain text, and multiple atomic facts are generated based on the semantic information, specifically including: Determining semantic information corresponding to each sentence contained in the domain text by using the information generation model and according to preset information generation instructions; According to the semantic information, a sentence structure type matching each atomic fact to be generated is determined, and each atomic fact is generated based on the sentence structure type.

[0009] Optionally, according to the feature code corresponding to the atomic fact, a target criterion matching the atomic fact is determined in a domain knowledge base of the domain to which the domain text belongs, specifically including: Determining the matching degree between each knowledge criterion and the atomic fact according to the feature code corresponding to the atomic fact and the feature code corresponding to each knowledge criterion in the domain knowledge base; Based on the matching degree, a target criterion matching the atomic fact is determined from among the knowledge criteria.

[0010] Optionally, according to the matching degree, determining a target criterion matching the atomic fact from among the knowledge criteria specifically includes: Determine the criterion whose matching degree with the atomic fact is higher than the preset matching degree as the target criterion, or The knowledge criteria are sorted in descending order of matching degree, and the knowledge criteria ranked before a specified position are determined as the target criteria.

[0011] Optionally, determining the confidence corresponding to each atomic fact through the confidence detection model specifically includes: For each atomic fact, if the atomic fact matches multiple target criteria, then determining a discrimination result obtained based on each target criterion through the confidence detection model based on a preset confidence detection instruction, wherein the discrimination result is used to determine whether the atomic fact is true; The confidence level corresponding to the atomic fact is determined according to the number of discrimination results that determine that the atomic fact is true and / or the number of discrimination results that determine that the atomic fact is not true.

[0012] Optionally, each atomic fact and each target criterion is input into a preset confidence detection model to determine the confidence corresponding to each atomic fact through the confidence detection model, specifically including: Input each atomic fact and each target criterion into multiple confidence detection models to obtain the confidence corresponding to each atomic fact output by each confidence detection model; According to the confidence level corresponding to each atomic fact, the comprehensive confidence level of the domain text is determined, specifically including: Determine the undetermined comprehensive confidence corresponding to each confidence detection model according to the confidence corresponding to each atomic fact output by each confidence detection model; The same comprehensive confidence with the largest number among the pending comprehensive confidences corresponding to the confidence detection models is determined as the final comprehensive confidence of the domain text.

[0013] This specification provides a text confidence detection device, comprising: The acquisition module is used to obtain the domain text to be detected; a generation module, configured to input the domain text into a preset information generation model, determine semantic information corresponding to each sentence contained in the domain text through the information generation model, and generate a plurality of atomic facts based on the semantic information, wherein each atomic fact is a complete declarative sentence describing an independent event that cannot be separated; A retrieval module is used to determine a target criterion matching each atomic fact in a domain knowledge base of the domain to which the domain text belongs based on a feature code corresponding to each atomic fact; A detection module, configured to input each atomic fact and each target criterion into a preset confidence detection model, so as to determine the confidence corresponding to each atomic fact through the confidence detection model; The determination module is used to determine the comprehensive confidence of the domain text according to the confidence corresponding to each atomic fact.

[0014] This specification provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, it implements the above-mentioned text confidence detection method.

[0015] This specification provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the above-mentioned text confidence detection method when executing the program.

[0016] At least one of the above technical solutions adopted in this specification can achieve the following beneficial effects: In the text confidence detection method provided in this specification, a domain text to be detected is obtained; the domain text is input into a preset information generation model to determine the semantic information corresponding to each sentence contained in the domain text through the information generation model, and multiple atomic facts are generated based on the semantic information, wherein each atomic fact is a complete declarative sentence describing an independent event that cannot be separated; based on the feature code corresponding to each atomic fact, a target criterion matching the atomic fact is determined in the domain knowledge base of the domain to which the domain text belongs; each atomic fact and each target criterion is input into a preset confidence detection model to determine the confidence corresponding to each atomic fact through the confidence detection model; and based on the confidence corresponding to each atomic fact, the comprehensive confidence of the domain text is determined.

[0017] As can be seen from the above method, this solution effectively solves the technical bottlenecks of traditional confidence detection, such as its strong reliance on manual annotation and single evaluation dimension, through the automated extraction of atomic facts and matching with domain knowledge bases. Specifically, it generates inseparable atomic fact units based on semantic parsing, and combines them with domain feature encoding to achieve precise matching of criteria, shifting the evaluation basis from subjective annotation to objective knowledge-driven. At the same time, the confidence fusion calculation model of multiple atomic facts breaks through the limitations of traditional single-metric evaluation, significantly improving detection accuracy in complex scenarios and achieving a qualitative improvement in comprehensive confidence credibility and scenario adaptability. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] The drawings described herein are used to provide a further understanding of this specification and constitute a part of this specification. The exemplary embodiments and descriptions of this specification are used to explain this specification and do not constitute an improper limitation of this specification. In the drawings: Figure 1 A flowchart of a text confidence detection method provided in this specification; Figure 2 A schematic diagram of a criterion search process provided in this specification; Figure 3 A schematic diagram of a crowd-investing process with a comprehensive confidence level provided in this specification; Figure 4 A schematic diagram of a text confidence detection device provided in this specification; Figure 5 This manual provides a corresponding Figure 1 Schematic diagram of electronic equipment. DETAILED DESCRIPTION

[0019] To make the objectives, technical solutions, and advantages of this specification more clear, the following will clearly and completely describe the technical solutions of this specification in conjunction with the specific embodiments of this specification and the corresponding drawings. Obviously, the embodiments described are only part of the embodiments of this specification, not all of the embodiments. Based on the embodiments in this specification, all other embodiments obtained by ordinary technicians in this field without making any creative efforts are within the scope of protection of this specification.

[0020] The current text confidence evaluation methods mainly include the following: Confidence is calculated using statistical methods, including term frequency (TF) and inverse document frequency (IDF). A higher term frequency indicates a greater importance of the term to the text. However, these methods often ignore the deeper semantics and context of the text, resulting in large errors for complex texts.

[0021] Using supervised or unsupervised learning, existing annotated data is used to train models to evaluate the credibility of text. For example, the Support Vector Machine (SVM) machine learning algorithm is used to convert text into numerical features through feature learning methods such as Word2Vec, and then output as category probabilities through a decision function. This method has very high requirements on text quality and the number of annotations, and the output cannot intuitively reflect the confidence level.

[0022] The confidence of domain text is directly judged by a general large language model. The general large language model has been pre-trained and uses the Softmax function to output the probability distribution of each category, that is, the confidence value. However, domain text is very professional and has scientific and logical reasoning. It is difficult to appear or be cited directly on the Internet. The probability distribution calculated by the general large model without pre-training on domain text and without obtaining domain knowledge is not interpretable.

[0023] Based on this, this specification provides a text confidence detection method. The following describes in detail the technical solutions provided by each embodiment of this specification in conjunction with the accompanying drawings.

[0024] Figure 1 The following is a flow chart of a text confidence detection method provided in this specification, which includes the following steps: S101: Obtain the domain text to be detected; S102: Input the domain text into a preset information generation model to determine the semantic information corresponding to each sentence contained in the domain text through the information generation model, and generate multiple atomic facts based on the semantic information, wherein each atomic fact is a complete declarative sentence describing an independent event that cannot be separated.

[0025] In this specification, the execution entity for executing a text confidence detection method can be a designated device such as a server. Of course, it can also be a client installed on devices such as mobile phones, tablets, laptops, and desktop computers. For the sake of convenience of description, the following will only take the server as the execution entity as an example to illustrate a text confidence detection method provided in this specification.

[0026] Among them, the server can obtain the field text to be detected. In actual applications, the scenarios for text detection can include a variety of scenarios. For example, in the field of financial risk control, the server can perform confidence detection on texts involving corporate financial reports, press releases and social media sentiment to identify potential market manipulation or false propaganda; in the medical field, medical research papers, patient electronic medical records and drug instructions can be tested to ensure the accuracy of information; in the legal field, contract terms, emails and recording transcripts can be reviewed to determine their authenticity; in terms of social media and news content review, false news, rumors and misleading information can be identified. In addition, the above-mentioned text confidence detection scenarios can also include checking and optimizing document content, screening model training samples, etc., which are not specifically limited in this manual.

[0027] When the text is too long, in order to avoid missing some content during subsequent semantic analysis due to information overload, the server can split the domain text into multiple independent text segments based on the preset length.

[0028] For example, the length of the above article may be 2,000 words. Paragraphs that are closely related and have the same core idea or theme can be divided into one text segment to avoid context discontinuity.

[0029] The server can then input each text segment into the information generation model.

[0030] The server can pre-set the prompt template of the information generation instruction. The information generation model can be a pre-trained language model (such as BERT, GPT, etc.). During this process, the information generation model can perform grammatical analysis and semantic extraction on the text fragments, extract the core information of each sentence, and convert the sentence into a standard sentence structure based on the extracted core information. The large language model can further perform semantic combination and deduplication operations on the standard sentence structure based on the contextual relationship of the sentence (including parallelism, causality, time, etc.), and finally generate each atomic fact.

[0031] The prompt template example for the above information generation instruction is as follows: “Break the following paragraph down into its individual atomic facts, using the example.

[0032] Here are some examples: [enter] Brazil is the world's largest coffee producer and exporter, with annual coffee production accounting for about 35% of the world's total production, ranking first in the world, followed by Vietnam and Indonesia. [Output] Brazil is the world's largest coffee producer; Brazil is the world's largest exporter of coffee; Brazil's annual coffee production accounts for about 35% of the world's total production; Vietnam is a coffee producer; Indonesia is a coffee-producing country.” In this specification, the sentence structure types of standard sentence structures may include standard triple structures such as subject-predicate-object, subject-copula-predicate, attributive-adverbial-complement, etc. The information generation model can generate instructions based on preset information, determine the semantic information corresponding to each sentence contained in the domain text, and then determine the sentence structure type that matches each atomic fact to be generated based on each semantic information, and generate each atomic fact based on the matched sentence structure type.

[0033] Each atomic fact is a complete statement describing an independent, indivisible event. As the fundamental unit of structured information, atomic facts achieve atomic decomposition and reorganization of semantics through standard triples, such as subject-verb-object and subject-copula-predicate. The information generation model, based on pre-set instructions, first parses the semantic information of the domain text, then matches the most appropriate sentence structure type for each atomic fact to be generated, ultimately generating a logically complete statement based on this structure.

[0034] The core value of this mechanism lies in its ability to transform unstructured text into machine-processable discrete fact units by standardizing sentence structure. For example, during the semantic parsing phase, the model can decompose the complex sentence "X was born in Y, and his father is Z" into two atomic facts—"X's birthplace is Y" and "X's father is Z." Each fact independently carries information about a single event. This decomposition eliminates the interference of modifiers in natural language, making key information explicit.

[0035] This technical framework achieves the transformation from free text to machine-readable knowledge through the synergy of sentence structure constraints and semantic decomposition. Its advantages are not only reflected in the accuracy of information extraction, but also lay the foundation for cross-domain knowledge integration and automated reasoning through standardized output formats.

[0036] S103: According to the feature code corresponding to each atomic fact, a target criterion matching the atomic fact is determined in the domain knowledge base of the domain to which the domain text belongs.

[0037] The server can first convert the atomic fact into a feature code using a pre-trained encoder model, and then search the domain knowledge base of the domain to which the current domain text belongs based on the feature code, thereby determining the target criterion that matches the atomic fact.

[0038] Specifically, the server can determine the matching degree (such as cosine similarity) between each knowledge criterion and the atomic fact based on the feature coding corresponding to the atomic fact and the feature coding corresponding to each knowledge criterion in the domain knowledge base, and then determine the target criterion that matches the atomic fact among each knowledge criterion based on the matching degree.

[0039] The server may determine a criterion whose matching degree with the atomic fact is higher than a preset matching degree as the target criterion. The preset matching degree may be set according to actual conditions, and this specification does not impose any specific limitation on this.

[0040] In addition, the server may also sort the knowledge criteria in descending order of matching degree and determine the knowledge criteria ranked before a specified order as the target criteria. For example, the first N knowledge criteria with the highest matching degree may be used as the target criteria.

[0041] In this way, the target criteria corresponding to each atomic fact can be obtained. For ease of understanding, this specification provides a schematic diagram of the criteria retrieval process, such as Figure 2 shown.

[0042] Among them, taking the atomic fact "Country S is the world's largest coffee producer" as an example, after converting it into a feature code, it can be further semantically searched in the domain knowledge base to obtain the target criteria related to it, such as target criterion 1: "For a long time, Country S has been ranked as the world's number one coffee producer", target criterion 2: "Country S's market share has dropped to the current 35%, but it is still the world's largest coffee producer", target criterion 3: "Country S still leads the coffee industry with a global output share of 37.4%."

[0043] Before this, the server needs to pre-build the domain knowledge base of each field. For each field, the server can obtain the domain knowledge information of the field and convert the domain knowledge information into a specified text format. Then, the server performs data processing on the domain knowledge information in the specified text format to obtain each knowledge criterion and determine the feature representation corresponding to each knowledge criterion. Then, based on the feature representation corresponding to each knowledge criterion, the server constructs the domain knowledge base of the field.

[0044] The process of building a domain knowledge base can include the following steps: (1) Domain knowledge collection: The server can collect factual materials on a certain field such as chemistry, biology, law, and medicine, including books, documents, research papers, etc., or download public data sets from professional websites and databases in the field to obtain domain knowledge information.

[0045] (2) Data preprocessing: Optical Character Recognition (OCR) technology is used to convert the original format of domain knowledge information (such as PDF) into text formats such as markdown and txt. The original PDF file generally contains text, pictures, tables, formulas, etc. To facilitate the subsequent reading of factual materials, pictures can be omitted during the OCR process.

[0046] The text recognized by OCR may have problems such as information redundancy and format errors. In order to improve the quality of the text, the text can be cleaned as follows: Remove unnecessary content, including headers and footers, page numbers, image tags, etc. Repair paragraphs, consecutive lines or words that are broken, resulting in incorrect paragraph formatting, and merge the corresponding lines or words; Unified format, such as the same time and date format, upper and lower case, etc.

[0047] Simplify the text, such as removing the reference list, preface, etc.

[0048] After data cleaning, the following post-processing is required: Metadata collection: extracting file-related metadata information including number, name, author, year, abstract, language, etc. to facilitate data search; Complete information and generate summary, keywords and other information.

[0049] (3) Build a domain knowledge vector library: The process of building a domain knowledge vector library can include the following steps: a. Text vectorization: To facilitate semantic retrieval and content generation, rather than just keyword matching, text data can be stored in a vector database. The server can leverage the vectorization capabilities of pre-trained language models (such as the BERT model) to accurately generate word embeddings for the entire text, thereby obtaining feature encodings of atomic facts.

[0050] b. Create an index: The server selects a database that can store and retrieve vectors (such as Milvus / FAISS). To facilitate subsequent vector retrieval, you can first create an index in the vector database and select the appropriate index type and parameters based on the actual data scale.

[0051] c. Vectorized data storage: The server can write the knowledge criteria into the corresponding vector field to ensure that the vector dimensions are consistent.

[0052] The core code example for building a knowledge base is as follows: "from transformers import BertTokenizer, BertModel import torch # Load the pre-trained model: tokenizer = BertTokenizer.from_pretrained('bert-base-uncased') model = BertModel.from_pretrained('bert-base-uncased') def text_to_vector(text): inputs = tokenizer(text, return_tensors="pt") outputs = model(**inputs) embeddings = outputs.last_hidden_state.mean(dim=300) # Average pooling generates text vectors return embeddings: text = "Brazil is the world's largest coffee producer" # Example vector = text_to_vector(text) #Import pymilvus library from pymilvus import Collection, FieldSchema, CollectionSchema,DataType, connections import numpy as np # Connect to the Milvus service (default address and port): connections.connect(alias="default", host="localhost", # Milvus server address: port="19530" # Milvus default port) # Define the field, assuming the vector dimension is 300: id_field = FieldSchema(name="id", dtype=DataType.INT64, is_primary=True) vector_field=FieldSchema(name="embedding",dtype=DataType.FLOAT_VECTOR, dim=300) # IVF index parameters: index_params = { "metric_type": "L2", # Use L2 distance metric "params": {"nlist": 1024} # nlist is the number of clusters} # Create an IVF index: collection.create_index(field_name="embedding", index_params=index_params) # Create Schema: schema = CollectionSchema(fields=[id_field, vector_field],description="text vector collection") # Create a Collection: collection = Collection(name="text_collection", schema=schema) vectors = [vector] # Insert data into the collection: insert_result = collection.insert(vectors). " S104: Inputting each atomic fact and each target criterion into a preset confidence detection model to determine the confidence corresponding to each atomic fact through the confidence detection model; S105: Determine the comprehensive confidence of the domain text according to the confidence corresponding to each atomic fact.

[0053] The server may input each atomic fact and each target criterion into a preset confidence detection model, thereby determining the confidence corresponding to each atomic fact through the confidence detection model based on a preset confidence detection instruction.

[0054] The prompt template example of the confidence detection instruction is as follows: "Please determine or infer the truth of the following statements based on the factual material provided in the attachments without reference to additional information.

[0055] Answer "support", "not support" or "not relevant" and explain the reason for your answer in the following format: Declarative Sentence 1 | [Your Answer] | [Explanation] Statement 2 | [Your Answer] | [Explanation] Declarative Sentence 3 | [Your Answer] | [Explanation] The following are the principles of judgment Support: Find or infer the plausibility of the statement based on the information provided Unsupported: Not "supported," either incorrectly stated or unverified Not relevant: The statement was not found in the information provided and cannot be confirmed. Here are some examples: Original text: "The three major nutrients (elements) of the human body are carbohydrates (sugars), proteins, and fats. These can be found in food and are used to sustain life." The statement "Carbohydrate is one of the three major nutrients for the human body" was supported; The statement “Protein cannot be obtained through food” is not supported.” Among them, the confidence corresponding to each atomic fact can be divided into three cases, namely 1, -1, and 0, corresponding to support, non-support, and irrelevant respectively. Then the server can determine the comprehensive confidence of the domain text based on the confidence corresponding to each atomic fact.

[0056] In this specification, the comprehensive confidence can be the confidence of the entire domain text, which can be expressed as:

[0057] Furthermore, in this specification, for each atomic fact, if the atomic fact matches multiple target criteria, the server can determine the discrimination result obtained based on each target criterion through the above-mentioned confidence detection model based on the preset confidence detection instruction, wherein the discrimination result is used to determine whether the atomic fact is true, and then determine the confidence corresponding to the atomic fact based on the number of discrimination results that determine that the atomic fact is true and / or the number of discrimination results that determine that the atomic fact is not true.

[0058] For example, the server may determine the confidence level of the atomic fact based on the proportion of the number of discrimination results that confirm that the atomic fact is true to the total number of discrimination results, wherein the number of discrimination results that confirm that the atomic fact is true is positively correlated with its confidence level; For another example, the server may also determine the confidence level according to the ratio of the number of determination results that determine that the atomic fact is not true to the total number, wherein the number of determination results that determine that the atomic fact is not true is negatively correlated with the confidence level.

[0059] Of course, the server may also determine the confidence level based on the ratio between the number of determination results that determine that the atomic fact is true and the number of determination results that determine that the atomic fact is false.

[0060] In this process, the confidence level can be a specific score. The more the number of judgment results that determine that the atomic fact is true, or the greater the proportion of the number of judgment results that determine that the atomic fact is true, the higher the confidence level of the atomic fact. The more the number of judgment results that determine that the atomic fact is false, or the greater the proportion of the number of judgment results that determine that the atomic fact is false, the lower the confidence level of the atomic fact.

[0061] In addition, to ensure the reliability of confidence, the server can input each atomic fact and each target criterion into multiple different confidence detection models to obtain the confidence corresponding to each atomic fact output by each confidence detection model, and determine the pending comprehensive confidence corresponding to the confidence detection model based on the confidence corresponding to each atomic fact output by each confidence detection model. Then, the comprehensive confidence corresponding to these detection models is crowd-voted, so that the most identical comprehensive confidence among the pending comprehensive confidences corresponding to each confidence detection model is determined as the final comprehensive confidence of the domain text. For ease of understanding, this manual provides a schematic diagram of the crowd-voting process of comprehensive confidence, as shown in the figure. Figure 3 shown.

[0062] For example, taking atomic fact 1, "Country S is the world's largest coffee producer," and the target criterion, "Country S has always been the world's largest coffee producer; its market share has decreased to its current 35%, but it remains the world's largest coffee producer; and it still leads the coffee industry with a 37.4% share of global production," the output of Model A is: The target criterion supports atomic fact 1—"it remains the world's largest coffee producer," directly supporting the claim that Country S is the world's largest coffee producer. The output of Model B is: The target criterion supports atomic fact 1—the provided criterion material clearly states that Country S is the world's largest coffee producer. The output of Model C is: The target criterion is irrelevant to atomic fact 1—it cannot be proven whether it is supported based on the existing criterion material.

[0063] From this, we can conclude that the number of output results supporting the atomic fact is the largest, so it can be determined that the content stated by the atomic fact is supported by the target criterion (confidence is 1) and is correct.

[0064] After determining the comprehensive confidence, the server can perform processing tasks for the domain text based on the comprehensive confidence.

[0065] As can be seen from the above methods, this solution breaks through the traditional confidence assessment's strong reliance on manual intervention and the limitations of evaluation dimensions by building an automated atomic fact extraction and knowledge base mapping mechanism. It innovatively uses semantic parsing to generate the smallest indivisible fact unit, combined with domain feature encoding to achieve precise alignment of criteria, upgrading the evaluation system from a subjective annotation model to a knowledge-driven architecture. At the same time, it establishes a multi-dimensional atomic fact confidence fusion calculation framework, effectively overcoming the shortcomings of traditional single-parameter evaluation, significantly enhancing detection accuracy in complex scenarios, and achieving a leap forward in credibility evaluation from a single indicator to systematic cognition.

[0066] The above is one or more implementations of the text confidence detection method of this specification. Based on the same idea, this specification also provides a corresponding text confidence detection device, such as Figure 4 shown.

[0067] Figure 4 A schematic diagram of a text confidence detection device provided in this specification, including: The acquisition module 401 is used to acquire the domain text to be detected; A generation module 402 is configured to input the domain text into a preset information generation model, determine semantic information corresponding to each sentence contained in the domain text through the information generation model, and generate multiple atomic facts based on the semantic information, wherein each atomic fact is a complete declarative sentence describing an independent event that cannot be separated; Retrieval module 403, configured to determine, based on the feature code corresponding to each atomic fact, a target criterion matching the atomic fact in the domain knowledge base of the domain to which the domain text belongs; A detection module 404 is configured to input each atomic fact and each target criterion into a preset confidence detection model to determine the confidence corresponding to each atomic fact through the confidence detection model; The determination module 405 is configured to determine the comprehensive confidence of the domain text according to the confidence corresponding to each atomic fact.

[0068] Optionally, the generating module 402 is specifically configured to split the domain text into a plurality of text segments with independent themes based on a preset length; and input each text segment into the information generation model.

[0069] Optionally, the generation module 402 is specifically used to determine the semantic information corresponding to each sentence contained in the domain text through the information generation model according to preset information generation instructions; determine the sentence structure type that matches each atomic fact to be generated based on the semantic information, and generate each atomic fact based on the sentence structure type.

[0070] Optionally, the retrieval module 403 is specifically used to determine the matching degree between each knowledge criterion and the atomic fact based on the feature coding corresponding to the atomic fact and the feature coding corresponding to each knowledge criterion in the domain knowledge base; based on the matching degree, determine the target criterion that matches the atomic fact among the knowledge criteria.

[0071] Optionally, the retrieval module 403 is specifically used to determine the criterion whose matching degree with the atomic fact is higher than a preset matching degree as the target criterion, or to sort the knowledge criteria in order of matching degree from large to small, and determine the knowledge criterion ranked before the specified position as the target criterion.

[0072] Optionally, the determination module 405 is specifically used to, for each atomic fact, if the atomic fact matches multiple target criteria, determine the discrimination result obtained based on each target criterion through the confidence detection model based on a preset confidence detection instruction, and the discrimination result is used to determine whether the atomic fact is established; determine the confidence corresponding to the atomic fact based on the number of discrimination results that determine that the atomic fact is established and / or the number of discrimination results that determine that the atomic fact is not established.

[0073] Optionally, the detection module 404 is specifically configured to input each atomic fact and each target criterion into a plurality of confidence detection models to obtain the confidence corresponding to each atomic fact output by each confidence detection model; The determination module 405 is specifically configured to determine the undetermined comprehensive confidence corresponding to each confidence detection model according to the confidence corresponding to each atomic fact output by each confidence detection model; The same comprehensive confidence with the largest number among the pending comprehensive confidences corresponding to the confidence detection models is determined as the final comprehensive confidence of the domain text.

[0074] This specification also provides a computer-readable storage medium, which stores a computer program that can be used to execute the above Figure 1 A text confidence detection method is provided.

[0075] This manual also provides Figure 5 The one shown corresponds to Figure 1 Schematic diagram of the electronic equipment. Figure 5 As mentioned above, at the hardware level, the electronic device includes a processor, an internal bus, a network interface, a memory and a non-volatile memory, and may also include other hardware required for the business. The processor reads the corresponding computer program from the non-volatile memory into the memory and then runs it to achieve the above Figure 1 Of course, in addition to software implementation, this specification does not exclude other implementation methods, such as logic devices or a combination of software and hardware, etc. That is to say, the execution subject of the following processing flow is not limited to each logic unit, but can also be hardware or logic devices.

[0076] Improvements to a technology can be clearly categorized as either hardware improvements (for example, improvements to circuit structures like diodes, transistors, and switches) or software improvements (improvements to process flows). However, with technological advancements, many process flow improvements can now be considered direct improvements to hardware circuit structures. Designers almost always create the corresponding hardware circuit structure by programming the improved process flow into the hardware circuit. Therefore, it cannot be said that a process flow improvement cannot be implemented using physical hardware modules. For example, a programmable logic device (PLD), such as a field programmable gate array (FPGA), is an integrated circuit whose logical function is determined by user programming. Designers can integrate a digital system onto a PLD by programming it themselves, without having to hire a chip manufacturer to design and manufacture a dedicated integrated circuit chip. Moreover, nowadays, instead of manually fabricating integrated circuit chips, this programming is mostly performed using software called a "logic compiler." This is similar to the software compilers used during program development. Before compilation, the original code must be written in a specific programming language, called a hardware description language (HDL). There are many types of HDL, including ABEL (Advanced Boolean Expression Language), AHDL (Altera Hardware Description Language), Confluence, CUPL (Cornell University Programming Language), HDCal, JHDL (Java Hardware Description Language), Lava, Lola, MyHDL, PALASM, and RHDL (Ruby Hardware Description Language). Currently, the most commonly used are VHDL (Very-High-Speed Integrated Circuit Hardware Description Language) and Verilog. Those skilled in the art will also understand that by simply programming the method flow in one of these hardware description languages and then programming it into an integrated circuit, a hardware circuit that implements the logic method flow can be easily obtained.

[0077] The controller can be implemented in any suitable manner. For example, the controller can take the form of a microprocessor or processor and a computer-readable medium storing computer-readable program code (e.g., software or firmware) executable by the (micro)processor, logic gates, switches, an application-specific integrated circuit (ASIC), a programmable logic controller, and an embedded microcontroller. Examples of controllers include, but are not limited to, the following microcontrollers: ARC 625D, Atmel AT91SAM, Microchip PIC18F26K20, and Silicone Labs C8051F320. The memory controller can also be implemented as part of the memory control logic. Those skilled in the art will also appreciate that, in addition to implementing the controller purely in computer-readable program code, the controller can also be implemented in the form of logic gates, switches, an application-specific integrated circuit, a programmable logic controller, an embedded microcontroller, etc. by logically programming the method steps. Therefore, such a controller can be considered a hardware component, and the means for implementing the various functions included therein can also be considered as structures within the hardware component. Alternatively, the means for implementing the various functions can be considered both a software module implementing the method and a structure within the hardware component.

[0078] The systems, devices, modules, or units described in the above embodiments may be implemented by computer chips or entities, or by products having certain functions. A typical implementation device is a computer. Specifically, the computer may be, for example, a personal computer, a laptop computer, a cellular phone, a camera phone, a smartphone, a personal digital assistant, a media player, a navigation device, an email device, a game console, a tablet computer, a wearable device, or a combination of any of these devices.

[0079] For the convenience of description, the above devices are described as being divided into various units according to their functions. Of course, when implementing this specification, the functions of each unit can be implemented in the same or multiple software and / or hardware.

[0080] Those skilled in the art will appreciate that the embodiments of this specification may be provided as methods, systems, or computer program products. Thus, this specification may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware. Furthermore, this specification may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0081] This specification is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of this specification. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0082] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0083] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0084] In a typical configuration, a processing device includes one or more processors (CPUs), input / output interfaces, network interfaces, and memory.

[0085] Memory may include non-permanent storage in a computer-readable medium, random access memory (RAM) and / or non-volatile memory in the form of read-only memory (ROM) or flash RAM. Memory is an example of a computer-readable medium.

[0086] Computer-readable media includes permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. The information can be computer-readable instructions, data structures, program modules, or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disc read-only memory (CD-ROM), digital versatile disc (DVD) or other optical storage, magnetic cassettes, magnetic disk storage or other magnetic storage devices, or any other non-transmission media that can be used to store information that can be accessed by a processing device. As defined herein, computer-readable media does not include transitory computer-readable media (transitory media), such as modulated data signals and carrier waves.

[0087] It should also be noted that the terms "comprises," "includes," or any other variations thereof are intended to encompass non-exclusive inclusion, such that a process, method, commodity, or apparatus that includes a series of elements includes not only those elements but also other elements not explicitly listed, or includes elements inherent to such process, method, commodity, or apparatus. In the absence of further limitations, an element defined by the phrase "comprises a ..." does not exclude the presence of other identical elements in the process, method, commodity, or apparatus that includes the element.

[0088] Those skilled in the art will appreciate that the embodiments of this specification may be provided as methods, systems, or computer program products. Therefore, this specification may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware. Furthermore, this specification may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0089] This specification may be described in the general context of computer-executable instructions, such as program modules, executed by a computer. Generally, program modules include routines, programs, objects, components, data structures, and the like that perform specific tasks or implement specific abstract data types. This specification may also be practiced in distributed computing environments where tasks are performed by remote processing devices connected through a communications network. In a distributed computing environment, program modules may be located in both local and remote computer storage media, including storage devices.

[0090] The various embodiments in this specification are described in a progressive manner. Similar parts between the various embodiments can be referred to in conjunction with each other. Each embodiment focuses on the differences between the other embodiments. In particular, the system embodiments are generally similar to the method embodiments, so the description is relatively simple. For relevant parts, refer to the description of the method embodiments.

[0091] The foregoing is merely an example of the present invention and is not intended to limit the present invention. Various modifications and variations are possible for those skilled in the art. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention are intended to be included within the scope of the claims of the present invention.

Claims

1. A text confidence detection method, characterized in that: include: Get the domain text to be detected; Inputting the domain text into a preset information generation model to determine semantic information corresponding to each sentence contained in the domain text through the information generation model, and generating multiple atomic facts based on the semantic information, wherein each atomic fact is a complete declarative sentence describing an independent event that cannot be separated; According to the feature code corresponding to each atomic fact, a target criterion matching the atomic fact is determined in the domain knowledge base of the domain to which the domain text belongs; Inputting each atomic fact and each target criterion into a preset confidence detection model to determine the confidence corresponding to each atomic fact through the confidence detection model; According to the confidence corresponding to each atomic fact, the comprehensive confidence of the domain text is determined.

2. The method according to claim 1, wherein Input the domain text into a pre-trained information generation model, specifically including: Based on a preset length, the domain text is split into multiple independent text segments; Each text segment is input into the information generation model.

3. The method according to claim 1, wherein The information generation model is used to determine the semantic information corresponding to each sentence contained in the domain text, and multiple atomic facts are generated based on the semantic information, specifically including: Determining semantic information corresponding to each sentence contained in the domain text by using the information generation model and according to preset information generation instructions; According to the semantic information, a sentence structure type matching each atomic fact to be generated is determined, and each atomic fact is generated based on the sentence structure type.

4. The method according to claim 1, wherein According to the feature code corresponding to the atomic fact, a target criterion matching the atomic fact is determined in the domain knowledge base of the domain to which the domain text belongs, specifically including: Determining the matching degree between each knowledge criterion and the atomic fact according to the feature code corresponding to the atomic fact and the feature code corresponding to each knowledge criterion in the domain knowledge base; Based on the matching degree, a target criterion matching the atomic fact is determined from among the knowledge criteria.

5. The method according to claim 4, wherein According to the matching degree, a target criterion matching the atomic fact is determined from among the knowledge criteria, specifically including: Determine the criterion whose matching degree with the atomic fact is higher than the preset matching degree as the target criterion, or The knowledge criteria are sorted in descending order of matching degree, and the knowledge criteria ranked before a specified position are determined as the target criteria.

6. The method according to claim 1, wherein The confidence level corresponding to each atomic fact is determined by the confidence level detection model, specifically including: For each atomic fact, if the atomic fact matches multiple target criteria, then the confidence detection model, based on a preset confidence detection instruction, determines a discrimination result obtained based on each target criterion, and the discrimination result is used to determine whether the atomic fact is true; The confidence level corresponding to the atomic fact is determined according to the number of discrimination results that determine that the atomic fact is true and / or the number of discrimination results that determine that the atomic fact is not true.

7. The method according to claim 1, wherein Inputting each atomic fact and each target criterion into a preset confidence detection model to determine the confidence corresponding to each atomic fact through the confidence detection model, specifically including: Input each atomic fact and each target criterion into multiple confidence detection models to obtain the confidence corresponding to each atomic fact output by each confidence detection model; According to the confidence level corresponding to each atomic fact, the comprehensive confidence level of the domain text is determined, specifically including: Determine the undetermined comprehensive confidence corresponding to each confidence detection model according to the confidence corresponding to each atomic fact output by each confidence detection model; The same comprehensive confidence with the largest number among the pending comprehensive confidences corresponding to the confidence detection models is determined as the final comprehensive confidence of the domain text.

8. A text confidence detection device, characterized in that: include: The acquisition module is used to obtain the domain text to be detected; a generation module, configured to input the domain text into a preset information generation model, determine semantic information corresponding to each sentence contained in the domain text through the information generation model, and generate a plurality of atomic facts based on the semantic information, wherein each atomic fact is a complete declarative sentence describing an independent event that cannot be separated; A retrieval module is used to determine a target criterion matching each atomic fact in a domain knowledge base of the domain to which the domain text belongs based on a feature code corresponding to each atomic fact; A detection module, configured to input each atomic fact and each target criterion into a preset confidence detection model, so as to determine the confidence corresponding to each atomic fact through the confidence detection model; The determination module is used to determine the comprehensive confidence of the domain text according to the confidence corresponding to each atomic fact.

9. A computer-readable storage medium, characterized in that The storage medium stores a computer program, and when the computer program is executed by a processor, the method according to any one of claims 1 to 7 is implemented.

10. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When executing the program, the processor implements the method according to any one of claims 1 to 7.

Images