30 results about "Single sentence" patented technology

The invention relates to the technical field of voice processing, can be applied to business scenes of financial science and technology, medical health and the like, and discloses a context information-based voice emotion recognition method, device, equipment and medium, which comprises the following steps: receiving an original voice stream and generating an independent voice segment, recognizing a text and determining a speaker role type, and extracting an acoustic feature index; and generating a preliminary emotion label, generating context information in combination with the historical dialogue text, and inputting the context information, the preliminary emotion label, the speaker role type and the acoustic feature index into a multi-modal fusion module to generate an emotion judgment result. According to the method, multi-modal fusion is realized on the basis of context information by combining voice, text and role information, so that the emotion change of each role can be accurately recognized and understood in a complex dialogue scene, the problems of large single sentence emotion judgment error and neglect of the context information in a traditional method are avoided, and the accuracy and stability of emotion recognition are effectively improved.

The application discloses a kind of main grid side operation mode single identification method and related device, method includes: based on preset semantic rule to the current operation mode single of main grid side carries out first preprocessing operation, obtains multiple semantic single sentence;According to first regular expression, action recognition is carried out to semantic single sentence, obtains action mode list;Second preprocessing operation is carried out to action mode list, obtains simplified action mode list;Word division operation is carried out to the sentence in simplified action mode list, obtains multiple segmentation words;Based on segmentation words and preset main grid side information, according to second regular expression, equipment identification is carried out, obtains equipment ID;Based on preset power grid topological relation, according to action mode list and equipment ID determine target action information.The application solves the technical problem that prior art lacks the text recognition scheme for mode single.

The present disclosure provides a natural language processing apparatus and a natural language processing method that may determine whether a user's speech command is a compound sentence or a complex sentence based on an output of a natural language understanding module, and when the user's speech command is a compound sentence or a complex sentence, recursively call the natural language understanding module, thereby executing all of the plurality of functions expressed as a single sentence.

This application discloses a corpus processing method and system for a large-scale model search engine, relating to the field of data processing technology. The corpus processing method for a large-scale model search engine includes: obtaining a target word segmentation result based on a first sentence; obtaining the semantic contribution degree corresponding to each word element based on the target word segmentation result; filtering out word elements in the target word segmentation result whose semantic contribution degree is lower than a first preset value to obtain an optimized word segmentation result; obtaining the priority of the first sentence based on the optimized word segmentation result; and determining whether to input the first sentence as valid corpus into the large-scale model based on the priority. This application overcomes the limitations of traditional word segmentation and filtering, accurately identifying high-quality, low-frequency corpus with reliable sources and core semantics, preventing it from being overwhelmed by high-frequency, low-quality information, and simultaneously eliminating false, low-quality corpus based on keyword stuffing from the source, thereby improving the quality of the input corpus for the large-scale model.

The invention belongs to the technical field of generative artificial intelligence, and provides a data depolarization and alignment enhancement method for a large language model, electronic equipment and a storage medium. The method comprises the steps of general text collection, single sentence disassembly, template matching, emotion labeling, object conflict group division, prejudice text and neutral text division, neutral dialogue model construction, prejudice model construction, synchronous answer generation, double-model output difference degree calculation, answer examination and neutral answer output. According to the method, through template matching, emotion labeling and conflict group division, complex cross bias detection is converted into objects, and emotion comparison statistical analysis is performed, so that the depolarization complexity is reduced, and the recognition efficiency of bias texts is improved; by constructing the neutral dialogue model and the prejudice model, error reference is provided, the process that traditional human preference alignment needs to collect a large number of personnel evaluations is avoided, and the training of the neutral dialogue model and the examination process during work are optimized.

The present invention discloses a method for sentiment analysis of conversational text based on deep learning, comprising the following steps: S1, classifying and labeling a data set; S2, normalizing the divided data set; S3, extracting features from the text using a hierarchical GRU model; S4, initializing the training parameters of the GRU model; S5, training the GRU model; S6, inputting a prediction sentence to obtain a training result. The present invention proposes a hierarchical GRU model, wherein the bottom layer is a bidirectional GRU model that extracts single sentence features, and the upper layer bidirectional GRU models context information to obtain interaction features between sentences; an attention mechanism is added to the hidden layer of the bidirectional GRU, and its output is fused with a single word or utterance embedding to strengthen the information of each word or utterance in the context embedding. The present invention uses a pre-trained model to obtain single sentence text features, which can effectively solve the problem of small database size.

The present disclosure relates to a method, device, storage medium and electronic device for sentence segmentation. The method comprises: obtaining target audio data; extracting speech recognition text corresponding to the target audio data and a first time period corresponding to each recognized character in the speech recognition text in the target audio data; performing speaker segmentation on the target audio data to obtain a second time period corresponding to each speech segment in the target audio data; and performing speaker segmentation on the speech recognition text according to the first time period corresponding to each recognized character and the second time period corresponding to each speech segment to obtain a sentence segmentation result. Thus, the speaker time period information and the time period corresponding to each character in the speech recognition text in the target audio data can be effectively utilized to perform speaker segmentation on the speech recognition text, to reasonably and effectively segment the speaker conversion, to avoid a single sentence segment containing speech content of multiple speakers, and to improve the sentence segmentation effect.

The invention discloses a multi-subject semantic attribute binding image generation method based on a diffusion model. The method comprises the steps of improving independent coding of a single sentence into segmentation of the sentence into a plurality of sub-sentences and independent coding of the sub-sentences; improving a denoising process and improving a cross attention mechanism; clear semantic representation is obtained by independently coding clauses, a denoising process is divided into a reconstruction branch and a generation branch, intermediate potential representation is spliced and fused to improve consistency and visual quality, and meanwhile, independent attention guidance is realized by combining an improved cross attention mechanism and mask division, so that feature aliasing is avoided, and the accuracy and the reliability of the system are improved. And the local precision and the space decoupling capability are enhanced. According to the multi-entity semantic modeling and image-text alignment method, a finer and more stable solution is provided while high efficiency and universality are ensured, and the potential in multi-entity semantic modeling and image-text alignment is shown.

The application provides a false keyword detection method, device, storage medium and program product. The method comprises: based on a semantic boundary, performing splitting processing on a to-be-detected text to obtain a plurality of sentence units; performing feature extraction processing on each sentence unit to obtain a single sentence feature vector corresponding to each sentence unit; based on the single sentence feature vector corresponding to each sentence unit, determining a sentence sequence level hidden feature corresponding to each sentence unit and a full-text global feature of the to-be-detected text; determining a contradiction degree feature vector corresponding to each sentence unit according to the sentence sequence level hidden feature of each sentence unit; and determining a false keyword corresponding to the to-be-detected text according to the full-text global feature of the to-be-detected text and the contradiction degree feature vector corresponding to each sentence unit. The method improves the reliability of long text false keyword detection.

A method, computer system, and a computer program product for text corpus concreteness modification are provided. A computer performs natural language processing to determine a concreteness level of a first individual sentence of a text corpus. The computer generates, based on the natural language processing, a proposed change of the individual sentence. The individual sentence with the proposed change includes a modified concreteness level and preserves a general meaning of the individual sentence. The computer transmits the proposed change for presentation of the proposed change.

The invention discloses a real-time online text violation recognition method and system and a storage medium, and belongs to the technical field of text recognition. The method comprises the following steps: collecting user behavior data in real time, and generating a user-level weak tag; comparing and learning the user-level weak tags to obtain a single sentence code and a single sentence risk assessment model; automatically distributing a pseudo label and an attention weight for each sentence by utilizing a pseudo label distribution rule, and training to obtain a lightweight sentence-level model; for an input real-time text, calculating a violation probability through a lightweight sentence-level model to obtain a risk index; according to the risk indexes and a preset risk threshold value, performing grading processing, blocking users whose risk indexes are in a high risk interval, performing secondary reasoning and random sampling on the users whose risk indexes are in a middle and low risk area, and then performing manual auditing; and iteratively updating the user-level weak label according to a user disposal result and newly added report data, and realizing self-evolution of each model. A large number of user-level tags can be obtained at low cost, and the detection rate is improved.

The invention discloses a lyric translation method and device, a storage medium and a program product, and relates to the technical field of computers. According to the method, the objective cue word corresponding to the lyric full text is utilized to restrain the translation, so that the translation keeps the sentiment basic tone of the original text, the metaphor or cultural image loss caused by literal translation is avoided, the cultural suitability is improved, and the lyric translation accuracy is improved; moreover, in combination with translation results of paragraph-level translation and sentence-level translation, paragraph breakage and semantic deviation caused by a sentence-by-sentence translation mode are avoided, and the consistency of repeated parts of refrain can be ensured; meanwhile, in the sentence-level translation process, multi-granularity context perception is realized by combining a paragraph subject and a single sentence context, and translation fragmentation is avoided; and the lyric translation accuracy and effect are improved.

The present invention relates to the technical field of seismic data processing, and particularly relates to a method and system for collecting seismic disaster situation data based on multi-source data. The present invention forms suspected Internet buzzword groups by any adjacent word groups in each single sentence; obtains an updated set of keyword groups according to the word group distribution of each suspected Internet buzzword group in the single sentence and the correlation between different word groups in the keyword group set; screens out effective disaster situation description single sentences according to the word group distribution of the updated set of keyword groups appearing in each single sentence and the word group correlation between the corresponding single sentence and the updated set of keyword groups; obtains the retained single sentences of the disaster situation description at the real-time moment for each earthquake area according to the number of effective disaster situation description single sentences in each earthquake area at different moments. By analyzing the accurate performance of disaster situation information during an earthquake, the present invention screens out effective disaster situation description single sentences, improving the effectiveness of seismic disaster situation data collection.

The invention discloses a named entity recognition large model training method and a named entity recognition method.The named entity recognition model comprises a BERT layer, an Encoder layer, an MLP layer and a CRF layer which are in signal connection in sequence, and the training method comprises the following steps that a training set is obtained, the training set comprises an original text and a similar text which is equal to the original text in semantics but different from the original text in sentence pattern, and the original text and the similar text are both single sentences; the original text and the similar text are sequentially input into a BERT layer, an Encoder layer generates token vectors of all basic units in the original text and the similar text, the token vectors are grouped according to entity types in the basic units, all the token vectors in each group are pulled to be close, and the token vectors between the groups are pulled to be far; and inputting the token vector corresponding to the original text into an MLP layer and a CRF layer for model parameter adjustment. According to the method, the introduction of noise is reduced, the entity representation has no offset, and the identification capability, robustness and generalization capability of the model to the entity type are enhanced.

This invention discloses a method and apparatus for processing conversation information, relating to the field of customer service technology. The method includes: extracting individual sentences from conversational statements generated by customer service; classifying the individual sentences using preset positive factor models and negative factor models; wherein the positive factor model is obtained by training a classification model through multiple positive clusters generated from positive key sentence clustering, and the negative factor model is obtained by training a classification model through multiple negative clusters generated from negative key sentence clustering; determining the target evaluation type of the individual sentence based on the evaluation types set by the classification results corresponding to the positive and negative factor models and the classification results of the individual sentences; and determining a service evaluation for customer service based on a preset evaluation system containing evaluation types and the target evaluation type of the individual sentences. This implementation can identify objective problems existing in conversational services based on the conversation and effectively improve the accuracy of conversation analysis.

The invention provides a sentence processing method and device, electronic equipment, a computer program product and a computer readable storage medium. The method comprises the following steps: for each first session single sentence, extracting a tail sentence of the first session single sentence to obtain a tail sentence corresponding to the first session single sentence; clustering the tail sentences of the plurality of first session single sentences based on the first similarity of the tail sentences of the plurality of first session single sentences in semantics to obtain a first tail sentence group; aiming at each tail sentence in the first tail sentence group, determining a second session single sentence corresponding to the tail sentence in the plurality of first session single sentences; and determining a third session single sentence with an incomplete tail sentence and a fourth session single sentence with a complete tail sentence from the second session single sentences corresponding to each tail sentence, and replacing the tail sentence of the third session single sentence with the tail sentence of the fourth session single sentence. According to the method and the device, under the condition of ensuring semantic consistency, the tail sentence of the session single sentence with the incomplete tail sentence can be automatically replaced, so that the usability of the session single sentence is improved.

This application discloses a data processing method, apparatus, device, storage medium, and program product, relating to the field of artificial intelligence technology. The data processing method includes acquiring service dialogue data between customer service representatives and users, wherein the service dialogue data includes at least two consecutive messages from the same role in both the customer service representative and the user role; inputting the service dialogue data and a first prompt word into a first large language model to obtain a standard dialogue corpus output by the first large language model; using the first prompt word to constrain the first large language model to semantically integrate the at least two consecutive messages from the same role, forming a single-sentence dialogue text of alternating question-and-answer between the two roles; and constructing a training dialogue corpus based on the results of business topic verification and semantic structure verification in the standard dialogue corpus. This enables the rapid construction of a training dialogue corpus, solving the problem of low training efficiency.

The application provides a fake news detection method and related equipment, and belongs to the technical field of natural language processing and information authenticity detection. The method comprises the following steps: preprocessing and segmenting an original input text to obtain a sentence set and score a candidate claim set obtained; converting the sentence into a single sentence declarative expression to obtain a normalized claim, and structurally constructing the normalized claim; simultaneously performing information retrieval according to the structural construction mode to obtain a candidate evidence, performing multidimensional consistency verification on the structured claim and the candidate evidence, and scoring to obtain a total verification score; and correcting the total verification score by using the matching degree of a news site and an IP address, and outputting a verification result. The application prepositions noise processing and fact positioning, adopts a structured claim for fine-grained verification, and corrects through cross-information source matching of an IP address and a news site, so that the accuracy, interpretability and identification ability for numerical / time tampering of fake news detection are improved.

A terminal (10) comprises: a determination unit (11) that, if label assessments are to be made for each sentence in a text comprising a plurality of sentences, determines a first threshold value for assessing that no label is to be assigned and a second threshold value for assessing that a label is to be assigned; a first assessment unit (12) that, on the basis of a candidate label score representing the degree to which each of a plurality of candidate labels can be derived from a single sentence subjected to labeling in the text, the first threshold value, and the second threshold value, makes a label assessment regarding the sentence subjected to labeling; and a second assessment unit (13) that, if assessment by the first assessment unit (12) is not possible, makes another label assessment of the sentence subjected to labeling, additionally using information on a sentence other than the sentence subjected to labeling in the text.

The present invention discloses a method and device for identifying topics in a dialogue system in combination with context information. The method comprises the following steps: obtaining a topic data set, constructing a slot library according to the relationship between the topics of multiple sentences in the topic data set and entities of each category in the sentences, wherein each row of slots in the slot library includes different topics and each entity of different categories corresponding thereto; inputting a current sentence of a dialogue to be identified into a trained single-sentence topic identification model, predicting a first topic and a corresponding predicted probability value; identifying all entities in the current sentence and all previous sentences, and establishing a dialogue slot table according to all entities and their categories; determining a second topic of the current sentence and a corresponding matching probability value in response to the number of successful matches of the dialogue slot table in the slot library; determining a second topic of the current sentence and a corresponding matching probability value in response to the fact that the topic corresponding to the larger one of the predicted probability value and the matching probability value is the first topic or the second topic, and avoiding loss of context information.

Embodiments of the present specification provide a question sentence generation method and device, wherein the question sentence generation method comprises: obtaining a to-be-processed text and a target answer corresponding to the to-be-processed text; marking a phrase in the to-be-processed text to obtain phrase marking information and a phrase structure diagram; inputting the to-be-processed text, the target answer, the phrase marking information and the phrase structure diagram into a question sentence generation model, wherein the question sentence generation model is used to generate a question sentence corresponding to the to-be-processed text and the target answer; and obtaining a target question sentence output by the question sentence generation model. The unstructured text data is processed, so that the input source is no longer only a single sentence or a dialogue flow, the phrase for generating the target question sentence is determined according to the phrase marking information and the phrase structure diagram, and on the basis of ensuring the accuracy of generating the target question sentence, the diversity of generating the target question sentence is improved.

The invention provides a campus spoofing behavior identification method and apparatus, and an electronic device. The method comprises the following steps: acquiring original audio data of a campus environment; segmenting the original audio data to obtain a plurality of single-sentence audio clips divided according to natural sentences and timestamp information corresponding to the single-sentence audio clips; performing multi-dimensional feature extraction on each single-sentence audio clip to obtain multi-dimensional feature data of each single-sentence audio clip; according to the timestamp information of each single sentence audio clip, integrating the multi-dimensional feature data into a time-sequenced dialogue representation text; and according to the dialogue representation text, performing campus spoofing behavior identification and strategy analysis to generate a campus spoofing behavior identification result and a corresponding behavior intervention strategy. According to the technical scheme, the audio dialogue is converted into the dialogue representation text fusing the emotion and the side language features, high-precision recognition and interpretable analysis of the campus spoofing behavior can be achieved based on deep reasoning of the dialogue representation text, and operation guidance is provided for educators.

The embodiment of the invention provides a parameter information extraction method and device and electronic equipment, and relates to the technical field of data extraction, and the method comprises the steps: obtaining a target contract to be subjected to parameter information extraction; determining target positioning information corresponding to a target contract category to which the target contract belongs; based on the target positioning information, positioning a paragraph containing parameter information of the target parameter in the target contract to obtain a target paragraph corresponding to the target parameter; obtaining a plurality of parameter feature sentences corresponding to the target parameter; analyzing whether a target single sentence matched with any parameter feature sentence corresponding to the target parameter exists in the single sentences of the target paragraph corresponding to the target parameter; and in response to the existence of the matched target single sentence, extracting parameter information of a target parameter in the target contract based on the target single sentence. Therefore, the extraction precision of the parameter information in the contract can be improved.

The application discloses a kind of knowledge graph construction methods based on fine-grained retrieval and reverse restoration self-error correction, specifically: first, standard reference case library is constructed, and standard reference vector is calculated.Then, the long text S to be measured is disassembled into sentences to be processed;Each sentence is vectorized, and the similarity score of its standard reference vector is calculated, and the first standard reference case of each sentence is screened;Through large language model, the non-standard logical relationship contained in each sentence is extracted and vectorized, the cosine similarity is calculated, and the candidate mapping is obtained by descending arrangement, after summarizing, double-feature reordering is carried out using multi-round greedy reordering algorithm, and the first standard reference case is screened out, and the single sentence is spliced into large language model, and triple extraction is carried out;Finally, the relationship triple set of all sentences constitutes knowledge graph.The application improves the precision and recall rate of triple extraction under the premise of avoiding redundancy.

The invention discloses an event argument extraction task data synthesis method and system based on a large language model, belongs to the technical field of event argument extraction, and solves the problems of event logic breakage, concentrated argument distribution, single synthesized data style and the like in the existing event argument extraction task which synthesizes data by using the large language model. The method comprises the following steps: determining an argument type to be contained in the to-be-synthesized original data according to an event type to which an event in the obtained to-be-synthesized original data belongs and an event argument contained in the corresponding event type, and constructing a cue word A to generate a single sentence only containing a single event; constructing a prompt word B to generate a paragraph text containing a plurality of events; generating a single sentence or paragraph based on the cue word C; filtering results with the formal problem, and constructing a prompt word D for marking or adding an event trigger word; constructing a cue word E for reconstruction; constructing a cue word F input large language model to generate diversified and multi-stylized output; and constructing a cue word G to obtain final synthetic data. The method is used for event argument extraction.

The application belongs to the technical field of generative artificial intelligence, and provides a data debiasing and alignment enhancement method for a large language model, an electronic device and a storage medium, the method comprising: general text collection, single sentence disassembly, template matching, sentiment labeling, object conflict group division, bias text and neutral text division, neutral dialogue model construction, bias model construction, synchronous answer generation, double model output difference calculation, answer review and neutral answer output; the application converts complex cross-bias detection into object and sentiment comparison statistical analysis through template matching, sentiment labeling and conflict group division, reduces the complexity of debiasing, and improves the recognition efficiency of biased text; by constructing a neutral dialogue model and a bias model, an error reference is provided, the process of collecting a large number of personnel evaluations for traditional human preference alignment is avoided, and the training and review process of the neutral dialogue model during work are optimized.

The application provides a segmentation method and device of a judgment document, an electronic equipment and a storage medium, wherein the method comprises the following steps: inputting a target judgment document into a Bert model to obtain the vectorization representation of each sentence in the target judgment document; segmenting the target judgment document based on the vectorization representation of each sentence to obtain a preliminary segmentation result; determining the context information of each sentence in the target judgment document based on the vectorization representation of each sentence, and adjusting the preliminary segmentation result based on the context information to obtain a final segmentation result of the target judgment document. By inputting the judgment document into the Bert model, the vectorization representation of each sentence is obtained, and the segmentation result is further optimized based on the context information, so that the segmentation is more in line with the actual structure of the judgment document. The final segmentation result not only considers the semantic information of a single sentence, but also further improves the accuracy of segmentation through the context relationship.

The present invention discloses a method, device, and medium for improving the generalization capability of a cross-modal image retrieval model. The method comprises: obtaining an image dataset, annotating the image data, and obtaining a large-scale image-text dataset with a single descriptive style; analyzing the large-scale image-text dataset with a single descriptive style to extract sentence templates with a single style; generating a set of sentence templates with diverse styles based on the single sentence templates; combining the set of sentence templates with diverse styles and using a template-based diversity enhancement strategy, annotating the image data again to obtain a large-scale image-text dataset with diverse descriptive styles; constructing and initializing a multimodal backbone network; and training the multimodal backbone network using a noise-aware masking strategy based on the large-scale image-text dataset. The present invention improves the generalization performance of a cross-modal image retrieval model and can be widely applied in the fields of image processing and recognition technology.

Embodiments of the present specification provide a method and device for malicious query intention recognition of a large model, the method comprising: obtaining a query sequence of a target user interacting with a large model, wherein the query sequence comprises a plurality of query sentences; for each query sentence, determining a single-sentence jump degree indicator according to a first similarity with an adjacent query sentence and a maximum similarity with the query sequence; determining a first indicator value reflecting logical coherence of the query sequence according to the jump degree indicator of each query sentence; dividing the query sequence into a plurality of topic clusters, any topic cluster comprising a plurality of continuous query sentences; determining a second indicator value reflecting an average follow-up depth of the query sequence according to at least the cluster size of each topic cluster; and determining whether the query sequence has a malicious query intention for the large model according to the first indicator value and the second indicator value. The effectiveness of malicious query intention recognition for a large model can be improved.

The application discloses a kind of extraction methods, equipment and medium of central event, comprising the following steps: by determining the similarity of each single sentence included in the text to be extracted with the title of the text to be extracted, the first weight of trigger word included in each single sentence and the second weight of network security entity included in each single sentence, the central sentence is determined according to the similarity, the first weight and the second weight of each single sentence, the trigger word contained in the central sentence is determined, the event type pointed by the central sentence is determined based on the trigger word, the central event is obtained by calculating the central sentence and event type through BiLSTM model and CRF model, the central sentence is determined by three dimensions, the extraction range is reduced, the interference of secondary event on the extraction of central sentence is reduced, the error existing in the mode of pipeline extraction of central event is reduced by BiLSTM model and CRF model, so as to improve the convenience and effectiveness of central event extraction task.