An intelligent quality inspection method and system based on deep learning

Abstract

The present invention provides an intelligent quality inspection method and system based on deep learning, wherein an embodiment of the present invention obtains text messages and voice data from multiple channels from a database to form a large-scale multi-source data set; pre-processes the multi-source data set based on dialect recognition rules to identify and mark dialect vocabulary; introduces a context analysis algorithm to perform context analysis on the pre-processed multi-source data set to obtain an intermediate data set; uses a natural language processing model based on deep learning to analyze the intermediate data set, extracts basic language features, and obtains a feature set to analyze whether the feature set contains preset information; the technical solution provided by the present invention improves the consistency and accuracy of data processing, reduces misjudgments and missed detections, reduces operating costs, quickly adapts to new changing trends, and provides an efficient, reliable and automated quality inspection solution for enterprises and institutions.

Description

Technical Field

[0001] The present invention relates to the field of data processing technology, and in particular to an intelligent quality inspection method and system based on deep learning. Background Art

[0002] In today's highly information-based society, enterprises and institutions are faced with the challenge of processing massive amounts of communication data. These data come not only from traditional text messages and voice calls, but also from a variety of emerging channels such as social media and instant messaging. As the amount of data continues to increase, how to efficiently and accurately extract valuable information from these multi-source heterogeneous data and ensure the quality and compliance of the information has become an urgent problem that needs to be solved. Existing quality inspection methods often have difficulties in data integration. The data formats from different sources are different and there is a lack of unified standards, which makes cross-channel data integration very complicated. In addition, there is insufficient recognition of dialects and local expressions. Different dialects or local vocabulary are used in different places, which brings huge challenges to natural language processing. Traditional methods are difficult to effectively identify and understand such non-standard languages. Simple keyword matching and other methods cannot fully capture the true meaning of the text, especially in the case of semantic ambiguity or polysemous words, which can easily lead to misjudgment. Relying on manually defined rule bases for preset word filtering is not only time-consuming and labor-intensive, but also slow to respond to new content, and prone to missed or wrong detection. Summary of the invention

[0003] The embodiments of the present invention provide an intelligent quality inspection method and system based on deep learning, which are used to solve the problem that existing quality inspection methods in the prior art often have difficulties in data integration, different data formats from different sources, and lack of unified standards, making cross-channel data integration very complicated.

[0004] In a first aspect, an embodiment of the present invention provides an intelligent quality inspection method based on deep learning, comprising:

[0005] Acquire SMS and voice data from multiple channels from the database to form a large-scale multi-source data set;

[0006] Preprocessing the multi-source data set based on dialect recognition rules to identify and mark dialect vocabulary;

[0007] Introduce context analysis algorithm to perform context analysis on preprocessed multi-source data sets to obtain intermediate data sets;

[0008] The intermediate data set is analyzed using a natural language processing model based on deep learning, basic language features are extracted, and a feature set is obtained to analyze whether the feature set contains preset information.

[0009] In a second aspect, an embodiment of the present application provides an intelligent quality inspection system based on deep learning, including:

[0010] The acquisition module is used to obtain SMS and voice data from multiple channels from the database to form a large-scale multi-source data set;

[0011] A preprocessing module, used for preprocessing the multi-source data set based on dialect recognition rules to identify and mark dialect words;

[0012] An analysis module is used to introduce a context analysis algorithm to perform context analysis on the preprocessed multi-source data set to obtain an intermediate data set;

[0013] The extraction module is used to analyze the intermediate data set using a natural language processing model based on deep learning, extract basic language features, and obtain a feature set to analyze whether the feature set contains preset information.

[0014] In a third aspect, an embodiment of the present invention provides a computing device, comprising a processor and a memory, wherein the memory stores a computer program, and the processor is configured to run the computer program to execute an intelligent quality inspection method based on deep learning as described in any one of the first aspects.

[0015] In a fourth aspect, an embodiment of the present invention provides a computer storage medium having computer program instructions stored thereon, wherein the computer program instructions, when executed by a processor, implement an intelligent quality inspection method based on deep learning as described in any one of the first aspects.

[0016] In an embodiment of the present invention, text messages and voice data from multiple channels are obtained from a database to form a large-scale multi-source data set; the multi-source data set is preprocessed based on dialect recognition rules to identify and mark dialect vocabulary; a context analysis algorithm is introduced to perform context analysis on the preprocessed multi-source data set to obtain an intermediate data set; the intermediate data set is analyzed using a natural language processing model based on deep learning to extract basic language features to obtain a feature set to analyze whether the feature set contains preset information. The technical solution provided by the present invention significantly improves the consistency and accuracy of data processing, reduces misjudgments and missed detections, reduces operating costs, and quickly adapts to new trends, by efficiently integrating multi-channel data, accurately identifying and marking dialect vocabulary, enhancing context understanding and sentiment analysis, and detecting preset information with high precision, and uses an adaptive learning framework to achieve intelligent and automated quality inspection. It can quickly adapt to new trends and provide enterprises and institutions with an efficient, reliable and automated quality inspection solution.

[0017] These and other aspects of the present invention will become more apparent from the following description of the embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0019] Figure 1 A flowchart of an intelligent quality inspection method based on deep learning provided by an embodiment of the present invention;

[0020] Figure 2 A schematic diagram of the structure of an intelligent quality inspection system based on deep learning provided by an embodiment of the present invention;

[0021] Figure 3 A schematic diagram of the structure of a computing device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0022] In order to enable those skilled in the art to better understand the solutions of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention.

[0023] In some of the processes described in the specification and claims of the present invention and the above-mentioned figures, multiple operations that appear in a specific order are included, but it should be clearly understood that these operations may not be executed in the order in which they appear in this article or executed in parallel. The serial numbers of the operations, such as 101, 102, etc., are only used to distinguish different operations, and the serial numbers themselves do not represent any execution order. In addition, these processes may include more or fewer operations, and these operations may be executed in sequence or in parallel. It should be noted that the descriptions of "first", "second", etc. in this article are used to distinguish different messages, devices, modules, etc., do not represent the order of precedence, and do not limit the "first" and "second" to be different types.

[0024] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present invention.

[0025] In the prior art, traditional quality inspection methods usually rely on simple keyword matching and manually defined rule bases. This static and inflexible method is difficult to effectively identify and understand non-standard languages, and is prone to misjudgment and missed detection. In addition, the lack of context understanding and sentiment analysis also makes the existing methods incapable of dealing with complex polysemous words and fuzzy semantics. Based on this, the present invention provides an intelligent quality inspection method based on deep learning, such as Figure 1 The specific implementation steps of this method are as follows:

[0026] Step 101, acquiring SMS and voice data from multiple channels from a database to form a large-scale multi-source data set;

[0027] Identify the data sources that need to be integrated, including but not limited to SMS platforms, phone recording systems, social media platforms, etc.;

[0028] Establish connections with these data sources to ensure that the latest communication data can be obtained from various channels in real time or regularly. Extract SMS and voice data from various data sources through API interfaces or other data transmission protocols. This step may involve different data formats, such as text files, audio files, etc.

[0029] For voice data, preliminary processing is first performed, such as converting voice into text (using voice-to-text technology), clearing noise and redundant information in the data, such as deleting invalid SMS records, removing background noise in voice data, etc.;

[0030] Check and correct errors in the data, such as inconsistent formats and timestamp errors, to ensure data quality. Convert data from different channels into a unified format. For SMS data, a unified encoding format is required. For voice data, a unified audio format (such as WAV, MP3, etc.) is required.

[0031] Ensure that all metadata fields such as timestamps and user identifiers are consistent. Store the cleaned and standardized data in a centralized database. This database can be a relational database (such as MySQL, PostgreSQL) or a non-relational database (such as MongoDB, Cassandra). Choose an appropriate data storage solution based on actual needs.

[0032] Design reasonable data structures and indexes to support efficient data query and retrieval, integrate data from different channels into a unified dataset that contains all relevant communication data and has good accessibility and scalability, and ensure that the dataset can support subsequent preprocessing steps such as dialect recognition, context analysis, etc.

[0033] Step 102, preprocessing the multi-source data set based on dialect recognition rules to identify and mark dialect words;

[0034] Convert SMS and voice data in multi-source datasets into text in a unified format, and use speech-to-text technology (such as automatic speech recognition ASR) to convert voice data into text form, so that all data exists in a consistent text format.

[0035] It uses predefined dialect vocabulary and grammar patterns to identify dialect words and expressions from uniformly formatted text. The dialect vocabulary contains common dialect words and their standard language equivalents, while the grammar patterns help the system understand the sentence structure and usage of a specific dialect.

[0036] Accurately identify and extract dialect words from texts through natural language processing techniques, such as regular expressions and dictionary matching;

[0037] Based on the identified dialect words and expressions, natural language processing technology is used to mark them, including marking the position of each dialect word and its corresponding standardized vocabulary in the text. The marking process can use named entity recognition (NER) technology to mark the dialect words as special entity types, so that these dialect words can be clearly distinguished and processed in subsequent processing.

[0038] Step 103, introducing a context analysis algorithm to perform context analysis on the preprocessed multi-source data set to obtain an intermediate data set;

[0039] By building the usage scenario of each dialect word or expression, combined with information such as conversation history, user background, timestamp and geographic location, we can more accurately understand the true meaning of the text. At the same time, we use a constantly updated database of slang and Internet language to identify and convert these non-standard expressions from the text into standard language, preserving their original meaning and emotional information.

[0040] Then, we use sentiment analysis algorithms to determine the emotional state of the text fragments, and use intent recognition algorithms to analyze the speaker's intentions, so as to fully understand the emotional color and actual meaning of the text. On this basis, we extract contextual features including emotional intensity, sense of humor and irony to further enrich the semantic information of the text.

[0041] Finally, the text that has undergone contextual analysis is integrated with its additional contextual features to form an intermediate data set. Duplicates, irrelevant information, and low-quality samples are removed through the data cleaning process to ensure the quality and consistency of the intermediate data set, providing a high-quality foundation for subsequent natural language processing.

[0042] Step 104, using a natural language processing model based on deep learning to analyze the intermediate data set, extract basic language features, obtain a feature set, and analyze whether the feature set contains preset information;

[0043] In this step, a multimodal sentiment analysis model is used to fuse and process text and speech data, and sentiment dictionaries, acoustic features, and contextual information are combined to extract vocabulary selection features, syntactic structure features, and sentiment color features. The extracted vocabulary selection features, syntactic structure features, and sentiment color features are integrated into a feature set, the performance strength of each feature is evaluated, the initial preset information detection threshold is set, and the threshold is dynamically adjusted according to the actual performance to improve the detection accuracy;

[0044] The feature set and dynamically adjusted thresholds are input into the adaptive learning framework, the model is trained through the machine learning algorithm, the parameters are iteratively optimized, the false alarm rate is reduced, and finally it is determined whether the feature set contains the preset information.

[0045] In summary, the embodiments of the present invention significantly improve the coverage, accuracy and automation level of intelligent quality inspection through multi-channel data integration, dialect recognition and labeling, context analysis and the application of deep learning models.

[0046] Based on this, the present invention provides a specific embodiment, which obtains SMS and voice data from multiple channels from a database to form a large-scale multi-source data set, which specifically includes the following steps:

[0047] Convert SMS and voice data from multiple source datasets into text in a unified format;

[0048] Convert SMS and voice data from multiple channels into a unified text format. For SMS data, this step is usually straightforward because it is already in text form. For voice data, it is necessary to transcribe it into text using automatic speech recognition (ASR) technology to ensure that all data exists in a consistent text format.

[0049] Using predefined dialect vocabulary and grammar patterns, it can identify dialect words and expressions from uniformly formatted texts.

[0050] The dialect vocabulary database contains common dialect words and their standard language equivalents. The grammatical pattern helps the system understand the sentence structure and usage of a specific dialect. It uses regular expressions, dictionary matching and other technologies to identify dialect words from the text, analyze the sentence structure based on the grammatical pattern, and identify expressions that conform to the dialect grammar.

[0051] Based on the dialect vocabulary and expressions, marking the identified dialect vocabulary using natural language processing technology;

[0052] Mark the location of each dialect word and its corresponding standardized word in the text, for example, by marking the dialect words with special tags and attaching their standard language counterparts. In addition, use named entity recognition (NER) technology to mark the dialect words as special entity types;

[0053] The embodiment of the present invention not only simplifies subsequent natural language processing tasks through the above steps, but also significantly improves the overall understanding and processing capabilities, especially when processing data containing a large amount of dialect expressions, it can significantly improve the accuracy and reliability of context analysis and preset information detection.

[0054] The present invention provides a specific embodiment, which introduces a context analysis algorithm to perform context analysis on a preprocessed multi-source data set to obtain an intermediate data set, and specifically includes the following steps:

[0055] Use contextual information and context features to build the usage scenario of each dialect word, including conversation history, user background, timestamp, and geographic location;

[0056] In this step, the usage scenario of each dialect word is constructed through contextual information such as conversation history, user background, timestamp, and geographic location. The rich contextual information is used to more accurately understand the true meaning of the text, thereby improving the accuracy of understanding dialect words and expressions and reducing misjudgments and missed detections.

[0057] Based on a dynamic database of slang and Internet language, searching the text for slang and Internet language that matches the dynamic database, wherein the dynamic database dynamically updates the slang and Internet language;

[0058] In this step, a constantly updated slang and Internet language database is used to search the text for slang and Internet terms that match the database. This dynamic database is regularly updated to include the latest slang and Internet buzzwords and to capture the latest non-standard language expressions in a timely manner;

[0059] For the identified slang and internet language, context-adaptive translation technology is used to convert them into standard language while retaining the original meaning and emotional information;

[0060] In this step, context-adaptive translation technology is applied to convert the identified slang and Internet language into standard language while retaining its original meaning and emotional information. This allows non-standard language to be converted into standard language without losing important information.

[0061] Use sentiment analysis and intent recognition algorithms to determine the emotional state and speaker intent contained in text snippets;

[0062] In this step, sentiment analysis algorithms are used to determine the emotional state of text fragments, such as positive, negative or neutral sentiment, and quantify its intensity. At the same time, intention recognition algorithms are used to analyze the speaker's intention, such as asking, stating or commanding. Sentiment analysis and intention recognition can fully understand the emotional color and actual meaning of the text and improve the ability to understand complex semantics.

[0063] Extracting contextual features from the context-analyzed data, the contextual features including: emotional intensity, sense of humor, and irony;

[0064] In this step, contextual features including emotional intensity, sense of humor, and irony are extracted from the contextually analyzed data. By extracting these features, we can more comprehensively understand and analyze the emotional color and nuances of the text, thereby better grasping the overall meaning of the text.

[0065] Integrate the context-analyzed text with its additional context features to form an intermediate dataset, and perform data cleaning to remove duplicates, irrelevant information, and low-quality samples;

[0066] In this step, the text that has undergone contextual analysis is integrated with its additional contextual features to form an intermediate data set. Then data cleaning is performed to remove duplicates, irrelevant information, and low-quality samples to ensure the quality and consistency of the data set. Through integration and cleaning, a high-quality intermediate data set is generated, which provides a solid foundation for subsequent natural language processing.

[0067] To summarize, the embodiment of the present invention introduces a context analysis algorithm to perform in-depth context analysis on the preprocessed multi-source data set to obtain an intermediate data set containing rich context information, which not only improves the accuracy of subsequent natural language processing, but also enhances the ability to understand complex semantics, thereby providing a solid foundation for the detection of preset information.

[0068] The present invention provides a specific embodiment, which uses a natural language processing model based on deep learning to analyze an intermediate data set, extract basic language features, and obtain a feature set to analyze whether the feature set contains preset information, which specifically includes the following steps:

[0069] Analyzing the intermediate data set using a multimodal sentiment analysis technology based on deep learning to obtain a feature set, wherein the feature set includes: vocabulary selection features, syntactic structure features, and sentiment color features;

[0070] Use a multimodal sentiment analysis model to extract basic language features that reflect user emotions, including vocabulary selection features, syntactic structure features, and emotional color features;

[0071] Among them, the vocabulary selection feature is extracted through word frequency statistics and sentiment vocabulary matching technology to reflect the vocabulary characteristics used by users; the syntactic structure feature identifies the structural features of sentences by performing syntactic analysis on the text; the emotional color feature determines the emotional tendency of the text by combining the vocabulary selection feature and the syntactic structure feature and using the sentiment analysis algorithm;

[0072] Based on the performance strength of each feature in the feature set, a dynamic threshold adjustment mechanism is used to perform evaluation processing and dynamically adjust the threshold of preset information detection;

[0073] Based on the performance strength of each feature in the feature set, the evaluation model is used for quantitative analysis to evaluate the contribution of each feature to the detection of preset information, and the feature strength evaluation result is obtained. According to the evaluation result and in combination with the preset information standard, the initial preset information detection threshold is set. Then, the initial preset information detection threshold and dynamic threshold adjustment mechanism are used to adjust the threshold of preset information detection in real time according to the actual performance of each feature in the feature set during the actual detection process, so as to improve the accuracy and adaptability of the detection and ensure that the preset information can be effectively identified in different situations;

[0074] Inputting the feature set and the threshold value of the preset information detection into an adaptive learning framework for iterative optimization to determine whether the feature set contains the preset information;

[0075] The feature set extracted by the multimodal sentiment analysis technology and the preset information detection threshold after dynamic adjustment are input into the adaptive learning framework. Based on these inputs, the machine learning algorithm in the adaptive learning framework is used to perform model training so that the model can learn to recognize the pattern of preset information. Through the iterative optimization mechanism of the adaptive learning framework, the model parameters are continuously adjusted, the detection threshold is optimized, the accuracy of preset information detection is improved, and the false alarm rate is reduced. Finally, the optimized model is used to perform preset information detection on the feature set to determine whether the feature set contains preset information;

[0076] In summary, the embodiments of the present invention extract rich basic language features from the intermediate data set, and achieve efficient and accurate detection of preset information through dynamic threshold adjustment and adaptive learning framework, thereby improving the automation level of quality inspection.

[0077] Based on this, the present invention also provides a specific embodiment, which analyzes the intermediate data set based on the multimodal sentiment analysis technology of deep learning to obtain a feature set, which specifically includes the following steps:

[0078] Using a multimodal sentiment analysis model based on deep learning, the text and voice data in the intermediate data set are fused and processed, and the basic language features that reflect the user's emotions are extracted through comprehensive analysis of sentiment dictionaries, acoustic features, and contextual information;

[0079] The sentiment dictionary is used to identify sentiment words in the text, and the acoustic features (such as pitch, speaking speed, etc.) are used to capture the emotional expression in the voice. The context information is also taken into consideration to more accurately understand the user's emotional background and extract the basic language features that reflect the user's emotions.

[0080] Step (d12) based on the fusion data, through word frequency statistics and sentiment vocabulary matching technology, analyze the vocabulary characteristics used by users, extract vocabulary selection features that can reflect the user's emotional tendencies;

[0081] Based on the fused data, the word frequency statistics and sentiment word matching technology are used to analyze the characteristics of the words used by users. The word frequency statistics can reveal the words that users use more frequently, while sentiment word matching can identify the words with emotional colors in the text by comparing with the predefined sentiment dictionary. These word selection characteristics can reflect the user's language habits and emotional tendencies, such as the frequency of positive or negative word use, so as to better understand the user's emotional state;

[0082] Use natural language processing technology to perform syntactic analysis on text data, identify the structural features of sentences, and obtain syntactic structural features;

[0083] Use natural language processing technology to perform syntactic analysis on text data, identify the structural features of sentences, identify the main components of sentences (such as subject, predicate, object), clause relationships, modifiers, etc. Through syntactic analysis, the structural features of sentences can be extracted. These features help understand the logical relationship and grammatical structure of sentences, and provide richer information for subsequent sentiment analysis;

[0084] Combining the vocabulary selection features and syntactic structure features, using sentiment analysis algorithms, determine the sentiment color features corresponding to the fused data;

[0085] Combine the extracted vocabulary selection features and syntactic structure features, and use sentiment analysis algorithms to determine the emotional color features corresponding to the fused data. Sentiment analysis algorithms will comprehensively consider vocabulary selection, sentence structure, and contextual information to determine the emotional tendency of the text fragment, such as positive, negative, or neutral;

[0086] In summary, the embodiments of the present invention use multimodal sentiment analysis technology based on deep learning to extract rich basic language features from the intermediate data set, including vocabulary selection features, syntactic structure features, and emotional color features. These features not only help to understand the user's emotional state, but also provide a solid foundation for subsequent preset information detection.

[0087] Based on this, the present invention provides a specific embodiment, which uses a dynamic threshold adjustment mechanism to perform evaluation processing based on the performance strength of each feature in the feature set, and dynamically adjusts the threshold of preset information detection, specifically including the following steps:

[0088] Based on the performance strength of each feature in the feature set, an evaluation model is used to perform quantitative analysis to evaluate the contribution of each feature to the detection of the preset information, thereby obtaining a feature strength evaluation result;

[0089] Through the evaluation model, each feature in the feature set (such as vocabulary selection features, syntactic structure features, and emotional color features) is quantitatively analyzed. The evaluation model will calculate the contribution of each feature to the detection of preset information based on the performance strength of each feature in the actual data, determine which features are most critical for identifying preset information, and provide a basis for subsequent threshold setting. For example, some emotional color features may have a higher weight when detecting negative emotions, while specific vocabulary selection features may be more important in identifying specific types of preset words;

[0090] According to the feature strength evaluation result and in combination with the preset information standard, an initial preset information detection threshold is set;

[0091] Based on the feature strength evaluation results obtained in the previous step, combined with the preset information standard, the initial preset information detection threshold is set. The preset information standard includes known preset words, expressions, and related emotional strengths. By comprehensively considering these standards and feature strength evaluation results, a reasonable initial threshold can be set. This threshold is used to determine whether a text segment contains preset information. For example, if the strength of a certain emotional color feature exceeds the threshold, the text segment may be marked as potential preset information;

[0092] Using the initial preset information detection threshold and dynamic threshold adjustment mechanism, the preset information detection threshold is adjusted according to the actual performance of each feature in the feature set during the actual detection process;

[0093] Continuously monitor the actual performance of each feature in the feature set, especially when new data is continuously input, and automatically adjust the threshold according to the feedback of the detection results. For example, if a certain type of feature is found to frequently misreport or miss reports in actual detection, the weight and threshold of the feature will be adjusted accordingly;

[0094] More specifically, the embodiment of the present invention further provides a calculation formula for a preset information detection threshold to describe how to dynamically adjust the threshold for preset information detection. The specific calculation formula is as follows:

[0095] ;

[0096] in, Indicates at a point in time Adjusted preset information detection threshold; is the initial preset information detection threshold set according to the preset information standard; is the adjustment coefficient that changes with time, which reflects the different degrees of influence of feature intensity on the final threshold in different time periods; It is a policy Features at time point The importance weight of the feature changes over time, reflecting the change of the importance of the feature over time; It is Features at time point Strength of performance; is a constant used to adjust Impact on threshold adjustment; is the relationship between all features at a point in time The interaction effect of is a constant used to adjust the contextual environment influence factor Impact on threshold adjustment; It's at the time contextual factors, such as user background, conversation history, timestamp, and geographic location, is the number of features in the feature set.

[0097] This formula takes into account the changing trends of features over time, the interactions between features, such as how certain features may jointly enhance or weaken the effect of intermittent information detection under certain conditions, and the impact of the contextual environment on the adjustment of the value, which can make the preset information detection more intelligent and adaptable.

[0098] The calculation formula of the adjustment coefficient that changes with time is as follows:

[0099] ;

[0100] ;

[0101] in, is the time-varying adjustment factor; is the initial adjustment coefficient, which reflects the basic effect of feature intensity on threshold adjustment without the influence of other factors; is the decay rate parameter, which controls how quickly the importance of features decreases over time; It is a dynamic adjustment item that changes over time and can reflect the impact of dynamic factors such as external environment and user behavior on the importance of features; is the number of factors considered in the dynamic adjustment term; It is The weight of a factor reflects the influence of the factor on the importance of the feature; It is Factors at time The degree of influence can be based on user feedback, social hot spots or other external factors.

[0102] This step introduces an index adjustment term and dynamic adjustments , making the preset information detection process more flexible and enhancing adaptability.

[0103] To sum up, the embodiments of the present invention utilize a dynamic threshold adjustment mechanism to perform evaluation and processing based on the performance strength of each feature in the feature set, dynamically adjust the threshold for detecting preset information, and improve the system's adaptability and the accuracy and reliability of detecting preset information.

[0104] The present invention provides a specific embodiment, wherein the feature set and the threshold value of the preset information detection are input into an adaptive learning framework for iterative optimization to determine whether the feature set contains the preset information, specifically comprising the following steps:

[0105] Input the feature set extracted by multimodal sentiment analysis technology and the dynamically adjusted preset information detection threshold into the adaptive learning framework;

[0106] The basic language features extracted through multimodal sentiment analysis technology (including vocabulary selection features, syntactic structure features and emotional color features) and the dynamically adjusted preset information detection threshold are input into the adaptive learning framework. The adaptive learning framework is a system that can continuously adjust and optimize model parameters according to the input data to improve the performance of the model.

[0107] Based on the input feature set and detection threshold, the model is trained using a machine learning algorithm in an adaptive learning framework, where the model learns to recognize patterns of preset information based on the data in the feature set and the preset information criteria;

[0108] In the adaptive learning framework, based on the input feature set and detection threshold, machine learning algorithms (such as deep neural networks, support vector machines, etc.) are used to train the model. According to the data in the feature set and the preset information standard, the model is learned to identify the pattern of preset information, so as to understand which feature combinations are most likely to indicate preset information and distinguish different types of preset content;

[0109] Through the iterative optimization mechanism of the adaptive learning framework, the parameters of the model are adjusted and the detection threshold is optimized to improve the accuracy of the preset information detection and reduce the false alarm rate;

[0110] Through back propagation and other methods, the weights and biases of the model are adjusted to make the model perform better on the training data. According to the performance of the model on the validation set, the threshold of the preset information detection is dynamically adjusted to find the best balance point, ensuring a high recall rate while maintaining a low false alarm rate. Through continuous feedback and adjustment, the accuracy and robustness of the model are gradually improved;

[0111] Using the optimized model, performing preset information detection on the feature set to determine whether the feature set contains the preset information;

[0112] The feature set of the text to be detected is input into the optimized model. The model will output the probability that the text segment contains the preset information. According to the optimized detection threshold, if the probability exceeds the threshold, the text segment is judged to contain the preset information; otherwise, it is judged as non-preset information.

[0113] Figure 2 A schematic diagram of the structure of an intelligent quality inspection system based on deep learning is provided for the embodiment of the present application, such as Figure 2 As shown, the system includes:

[0114] An acquisition module 21 is used to acquire SMS and voice data from multiple channels from a database to form a large-scale multi-source data set;

[0115] A preprocessing module 22, configured to preprocess the multi-source data set based on a dialect recognition rule to recognize and mark dialect words;

[0116] An analysis module 23 is used to introduce a context analysis algorithm to perform context analysis on the preprocessed multi-source data set to obtain an intermediate data set;

[0117] The extraction module 24 is used to analyze the intermediate data set using a natural language processing model based on deep learning, extract basic language features, obtain a feature set, and analyze whether the feature set contains preset information.

[0118] Figure 2 The intelligent quality inspection system based on deep learning can perform Figure 1 The implementation principle and technical effects of the intelligent quality inspection method based on deep learning described in the illustrated embodiment will not be described in detail. The specific manner in which each module and unit performs operations in the intelligent quality inspection system based on deep learning in the above embodiment has been described in detail in the embodiment of the method, and will not be elaborated here.

[0119] Figure 2 An intelligent quality inspection system based on deep learning in the embodiment shown can be implemented as a computing device, such as Figure 3 As shown, the computing device may include a storage component 31 and a processing component 32;

[0120] The storage component 31 stores one or more computer instructions, wherein the one or more computer instructions are called and executed by the processing component 32 .

[0121] The processing component 32 is used to: obtain text messages and voice data from multiple channels from a database to form a large-scale multi-source data set; pre-process the multi-source data set based on dialect recognition rules to identify and mark dialect vocabulary; introduce a context analysis algorithm to perform context analysis on the pre-processed multi-source data set to obtain an intermediate data set; use a natural language processing model based on deep learning to analyze the intermediate data set, extract basic language features, and obtain a feature set to analyze whether the feature set contains preset information.

[0122] The processing component 32 includes one or more processors to execute computer instructions to complete all or part of the steps in the above method. Of course, the processing component can also be implemented by one or more application-specific integrated circuits (AICs), digital signal processors (DPs), digital signal processing devices (DPDs), programmable logic devices (PLDs), field programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors or other electronic components to perform the above method.

[0123] The storage component 31 is configured to store various types of data to support operations at the terminal. The storage component can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (RAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk or optical disk.

[0124] Computing devices also include other components, such as input / output interfaces, display components, and communication components.

[0125] The input / output interface provides an interface between the processing component and the peripheral interface module, and the above-mentioned peripheral interface module can be an output device or an input device.

[0126] The communication component is configured to facilitate, among other things, wired or wireless communications between the computing device and other devices.

[0127] Among them, the computing device can be a physical device or an elastic computing host provided by a cloud computing platform, etc. In this case, the computing device can refer to a cloud server, and the above-mentioned processing components, storage components, etc. can be basic server resources rented or purchased from the cloud computing platform.

[0128] The embodiment of the present invention further provides a computer storage medium storing a computer program, which can achieve the above-mentioned Figure 1 An intelligent quality inspection method based on deep learning in the illustrated embodiment.

[0129] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0130] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0131] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0132] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. An intelligent quality inspection method based on deep learning, characterized in that: include: Acquire SMS and voice data from multiple channels from the database to form a large-scale multi-source data set; Preprocessing the multi-source data set based on dialect recognition rules to identify and mark dialect vocabulary; Introduce context analysis algorithm to perform context analysis on preprocessed multi-source data sets to obtain intermediate data sets; Use a natural language processing model based on deep learning to analyze the intermediate data set, extract basic language features, and obtain a feature set to analyze whether the feature set contains preset information; Among them, the natural language processing model based on deep learning is used to analyze the intermediate data set, extract basic language features, and obtain a feature set to analyze whether the feature set contains preset information, including: Analyzing the intermediate data set using a multimodal sentiment analysis technology based on deep learning to obtain a feature set, wherein the feature set includes: vocabulary selection features, syntactic structure features, and sentiment color features; Based on the performance strength of each feature in the feature set, a dynamic threshold adjustment mechanism is used to perform evaluation processing and dynamically adjust the threshold of preset information detection; Inputting the feature set and the threshold value of the preset information detection into an adaptive learning framework for iterative optimization to determine whether the feature set contains the preset information; The feature set and the threshold value of the preset information detection are input into the adaptive learning framework for iterative optimization to determine whether the feature set contains the preset information, including: Input the feature set extracted by multimodal sentiment analysis technology and the dynamically adjusted preset information detection threshold into the adaptive learning framework; Based on the input feature set and detection threshold, the model is trained using a machine learning algorithm in an adaptive learning framework, where the model learns to recognize patterns of preset information based on the data in the feature set and the preset information criteria; Through the iterative optimization mechanism of the adaptive learning framework, the parameters of the model are adjusted and the detection threshold is optimized to improve the accuracy of the preset information detection and reduce the false alarm rate; The optimized model is used to perform preset information detection on the feature set to determine whether the feature set contains the preset information.

2. The intelligent quality inspection method based on deep learning as claimed in claim 1, wherein the step of preprocessing the multi-source data set based on dialect recognition rules for identifying and marking dialect vocabulary comprises: Convert SMS and voice data from multiple source datasets into text in a unified format; Using predefined dialect vocabulary and grammar patterns, it can identify dialect words and expressions from uniformly formatted texts. Based on the dialect vocabulary and expressions, the identified dialect vocabulary is marked using natural language processing technology.

3. The intelligent quality inspection method based on deep learning as claimed in claim 1, wherein the step of introducing a context analysis algorithm to perform context analysis on the preprocessed multi-source data set to obtain an intermediate data set comprises: Use contextual information and context features to build the usage scenario of each dialect word, including conversation history, user background, timestamp, and geographic location; Based on a dynamic database of slang and Internet language, searching the text for slang and Internet language that matches the dynamic database, wherein the dynamic database dynamically updates the slang and Internet language; For the identified slang and internet language, context-adaptive translation technology is used to convert them into standard language while retaining the original meaning and emotional information; Use sentiment analysis and intent recognition algorithms to determine the emotional state and speaker intent contained in text snippets; Extracting contextual features from the context-analyzed data, the contextual features including: emotional intensity, sense of humor, and irony; The context-analyzed text is integrated with its additional contextual features to form an intermediate dataset, and data cleaning is performed to remove duplicates, irrelevant information, and low-quality samples.

4. The intelligent quality inspection method based on deep learning according to claim 1, using a multimodal sentiment analysis technology based on deep learning to analyze the intermediate data set to obtain a feature set, including: Using a multimodal sentiment analysis model based on deep learning, the text and voice data in the intermediate data set are fused and processed, and the basic language features that reflect user emotions are extracted through comprehensive analysis of sentiment dictionaries, acoustic features, and contextual information. Based on the fused data, the word frequency statistics and sentiment word matching technology are used to analyze the vocabulary characteristics used by users and extract the vocabulary selection features that can reflect the user's emotional tendencies; Use natural language processing technology to perform syntactic analysis on text data, identify the structural features of sentences, and obtain syntactic structural features; Combining the vocabulary selection features and syntactic structure features, the sentiment analysis algorithm is used to determine the emotional color features corresponding to the fused data.

5. The intelligent quality inspection method based on deep learning as claimed in claim 1, based on the performance strength of each feature in the feature set, using a dynamic threshold adjustment mechanism to perform evaluation processing and dynamically adjust the threshold of preset information detection, including: Based on the performance strength of each feature in the feature set, an evaluation model is used to perform quantitative analysis to evaluate the contribution of each feature to the detection of the preset information, thereby obtaining a feature strength evaluation result; According to the feature strength evaluation result and in combination with the preset information standard, an initial preset information detection threshold is set; The initial preset information detection threshold and the dynamic threshold adjustment mechanism are used to adjust the preset information detection threshold according to the actual performance of each feature in the feature set during the actual detection process.

6. An intelligent quality inspection system based on deep learning, characterized in that: include: The acquisition module is used to obtain SMS and voice data from multiple channels from the database to form a large-scale multi-source data set; A preprocessing module, used for preprocessing the multi-source data set based on dialect recognition rules to identify and mark dialect words; An analysis module is used to introduce a context analysis algorithm to perform context analysis on the preprocessed multi-source data set to obtain an intermediate data set; An extraction module is used to analyze the intermediate data set using a natural language processing model based on deep learning, extract basic language features, obtain a feature set, and analyze whether the feature set contains preset information; Among them, the natural language processing model based on deep learning is used to analyze the intermediate data set, extract basic language features, and obtain a feature set to analyze whether the feature set contains preset information, including: Analyzing the intermediate data set using a multimodal sentiment analysis technology based on deep learning to obtain a feature set, wherein the feature set includes: vocabulary selection features, syntactic structure features, and sentiment color features; Based on the performance strength of each feature in the feature set, a dynamic threshold adjustment mechanism is used to perform evaluation processing and dynamically adjust the threshold of preset information detection; Inputting the feature set and the threshold value of the preset information detection into an adaptive learning framework for iterative optimization to determine whether the feature set contains the preset information; The feature set and the threshold value of the preset information detection are input into the adaptive learning framework for iterative optimization to determine whether the feature set contains the preset information, including: Input the feature set extracted by multimodal sentiment analysis technology and the dynamically adjusted preset information detection threshold into the adaptive learning framework; Based on the input feature set and detection threshold, the model is trained using a machine learning algorithm in an adaptive learning framework, where the model learns to recognize patterns of preset information based on the data in the feature set and the preset information criteria; Through the iterative optimization mechanism of the adaptive learning framework, the parameters of the model are adjusted and the detection threshold is optimized to improve the accuracy of the preset information detection and reduce the false alarm rate; The optimized model is used to perform preset information detection on the feature set to determine whether the feature set contains the preset information.

7. An electronic device comprising a memory and a processor, characterized in that: A computer program is stored in the memory, and the processor is configured to run the computer program to execute the deep learning-based intelligent quality inspection method according to any one of claims 1 to 5.

8. A readable storage medium, characterized in that: The readable storage medium stores a computer program, and when the computer program is executed by a computer, it implements the intelligent quality inspection method based on deep learning as described in any one of claims 1 to 5.

Images