Text data processing method and device, and electronic device

Abstract

The present disclosure provides a text data processing method and device, electronic equipment and storage medium, relates to the technical field of artificial intelligence, in particular to the technical field of data processing, deep learning and the like. The specific implementation scheme is: obtaining a first text data set to be converted in a text data set; converting the first text data set into an entity data set; in response to a quality evaluation result of entity data in the entity data set being greater than a quality evaluation threshold, labeling a second text data set to be labeled in the text data set by using the entity data set to obtain a labeled data set; and outputting the labeled data set.

Description

Technical Field

[0001] This disclosure relates to the field of artificial intelligence technology, and in particular to the fields of data processing and deep learning. Background Technology

[0002] Currently, sequence labeling techniques are commonly used to process text data. However, this method requires a large amount of high-quality labeled data to train the mention recognition model, and manually labeling a large amount of high-quality labeled data would lead to excessive model development costs. Summary of the Invention

[0003] This disclosure provides a method, apparatus, electronic device, and storage medium for processing text data.

[0004] According to one aspect of this disclosure, a method for processing text data is provided. The method may include: acquiring a first text dataset to be converted from a text dataset; converting the first text dataset into an entity dataset; in response to a quality assessment result of entity data in the entity dataset exceeding a quality assessment threshold, labeling a second text dataset to be labeled using the entity dataset to obtain a labeled dataset, wherein the labeled dataset includes the entity dataset; and outputting the labeled dataset, wherein the labeled dataset is used to obtain an entity recognition model through supervised training, and the entity recognition model is used to recognize entity data in the text to be recognized.

[0005] According to another aspect of this disclosure, a method for determining a model is also provided. This method may include: acquiring a labeled dataset, wherein the labeled dataset is obtained by labeling a text dataset to be labeled using an entity dataset, and includes an entity dataset, wherein the quality assessment result of the entity data in the entity dataset is greater than a quality assessment threshold, and the entity dataset is obtained by converting a text dataset to be converted from the text dataset; and performing supervised training based on the labeled dataset to obtain an entity recognition model, wherein the entity recognition model is used to recognize entity data in the text to be recognized.

[0006] According to one aspect of this disclosure, a text data processing apparatus is provided. The apparatus may include: a first acquisition unit for acquiring a first text dataset to be converted from a text dataset; a conversion unit for converting the first text dataset into an entity dataset; an annotation unit for annotating a second text dataset to be annotated from the text dataset using the entity dataset in response to a quality assessment result of entity data in the entity dataset exceeding a quality assessment threshold, thereby obtaining an annotated dataset, wherein the annotated dataset includes the entity dataset; and an output unit for outputting the annotated dataset, wherein the annotated dataset is used to obtain an entity recognition model through supervised training, and the entity recognition model is used to recognize entity data in the text to be recognized.

[0007] According to one aspect of this disclosure, a model determination apparatus is provided. The apparatus may include: a second acquisition unit for acquiring a labeled dataset, wherein the labeled dataset is obtained by labeling a text dataset to be labeled using an entity dataset, and includes an entity dataset, wherein the quality assessment result of the entity data in the entity dataset is greater than a quality assessment threshold, and the entity dataset is obtained by converting a text dataset to be converted from the text dataset; and a training unit for performing supervised training based on the labeled dataset to obtain an entity recognition model, wherein the entity recognition model is used to recognize entity data in the text to be recognized.

[0008] According to another aspect of this disclosure, an electronic device is provided, comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform a text data processing method according to embodiments of this disclosure.

[0009] According to another aspect of this disclosure, a non-transitory computer-readable storage medium storing computer instructions is provided, wherein the computer instructions are used to cause a computer to perform a text data processing method according to embodiments of this disclosure.

[0010] According to another aspect of this disclosure, a computer program product is provided, including a computer program that, when executed by a processor, implements a text data processing method according to embodiments of this disclosure.

[0011] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of this disclosure, nor is it intended to limit the scope of this disclosure. Other features of this disclosure will become readily apparent from the following description. Attached Figure Description

[0012] The accompanying drawings are provided to better understand this solution and do not constitute a limitation of this disclosure. Wherein:

[0013] Figure 1 This is a flowchart of a text data processing method according to an embodiment of the present disclosure;

[0014] Figure 2(a) is a flowchart of a method for determining a model according to an embodiment of the present disclosure;

[0015] Figure 2(b) is a schematic diagram of a computer product for performing a method for determining a model according to an embodiment of the present disclosure;

[0016] Figure 2(c) is a schematic diagram of a human-computer interaction interface for performing a model determination method according to an embodiment of the present disclosure;

[0017] Figure 3 This is a schematic diagram of a semi-supervised mention mining system according to an embodiment of the present disclosure;

[0018] Figure 4 This is a schematic diagram of a model framework for a quality assessment model according to an embodiment of the present disclosure;

[0019] Figure 5 This is a schematic diagram of a text data processing apparatus according to an embodiment of the present disclosure;

[0020] Figure 6 This is a schematic diagram of a model determining device according to an embodiment of the present disclosure;

[0021] Figure 7 This is a schematic diagram of an electronic device according to an embodiment of the present disclosure of a text data processing method. Detailed Implementation

[0022] The exemplary embodiments of this disclosure are described below with reference to the accompanying drawings, including various details of the embodiments to aid understanding, and should be considered merely exemplary. Therefore, those skilled in the art will recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of this disclosure. Similarly, for clarity and brevity, descriptions of well-known functions and structures are omitted in the following description.

[0023] The following describes a method for processing text data according to an embodiment of this disclosure.

[0024] Figure 1 This is a flowchart of a text data processing method according to an embodiment of the present disclosure, such as... Figure 1 As shown, the method may include the following steps:

[0025] Step S102: Obtain the first text dataset to be converted from the text dataset.

[0026] In the technical solution provided in step S102 of this disclosure, the aforementioned text dataset may include internally produced document data and internal enterprise document data. For example, the text dataset may include purchase orders, industry reports, business attachments, sales contracts, employment agreements, commercial invoices, resumes, etc. The text dataset includes a first text dataset, which may be a collection of text data to be converted within the text dataset.

[0027] It should be noted that the text types in the above-mentioned text dataset are merely illustrative examples of embodiments of the present invention, and do not specifically limit the text types in the text dataset. Any text in any text dataset is within the protection scope of the embodiments of this disclosure, and will not be listed here.

[0028] In this embodiment, a first text dataset to be transformed can be obtained from the text dataset, wherein the first text dataset can be a collection of text data to be mined.

[0029] Step S104: Convert the first text dataset into an entity dataset.

[0030] In the technical solution provided by step S104 of this disclosure, after obtaining the first text dataset to be converted from the text dataset, the first text dataset is converted into an entity dataset. The entity dataset can be a mined mention dataset, which can include words or phrases. A mention can be defined as a reference to an entity in natural language text or a language fragment expressing an entity in natural text. The entity can be a named entity, a nominal entity, or a pronominal entity, without specific limitations here.

[0031] In this embodiment, when converting the first text dataset into an entity dataset, an unsupervised mining method or a semi-supervised mining method can be used. The unsupervised mining method can be high-frequency co-occurrence phrase mining, part-of-speech template mining, or dependency relation template mining. The entity dataset can be used for tasks such as information extraction, retrieval, classification modeling, and topic analysis, without specific limitations here.

[0032] For example, when converting the first text dataset into an entity dataset, a high-frequency co-occurrence phrase mining method can be used to identify the most frequent mentions in the first text dataset as entity datasets, or a part-of-speech template mining method can be used to identify the mentions in the first text dataset that match the part-of-speech template as entity datasets, or a dependency relation template mining method can be used to identify the mentions in the first text dataset that match the dependency relation template as entity datasets.

[0033] It should be noted that the above method for converting the first text dataset into an entity dataset is only an example. There is no limitation on the method for converting the first text dataset into an entity dataset. Any method and process for converting the first text dataset into an entity dataset is within the protection scope of this disclosure, and will not be elaborated here.

[0034] Step S106: In response to the quality assessment result of the entity data in the entity dataset being greater than the quality assessment threshold, the entity dataset is used to annotate the second text dataset to be labeled in the text dataset to obtain the labeled dataset, wherein the labeled dataset includes the entity dataset.

[0035] In the technical solution provided by step S106 of this disclosure, after converting the first text dataset into an entity dataset, it can be determined whether the quality assessment result of the entity data in the entity dataset is greater than the quality assessment threshold. Then, in response to the fact that the quality assessment result of the entity data in the entity dataset is greater than the quality assessment threshold, the entity dataset can be used to annotate the second text dataset to be labeled in the text dataset to obtain the labeled dataset.

[0036] In this embodiment, a mention quality assessment model can be used to assess the quality of entity data in the entity dataset. If the quality assessment result is less than the quality assessment threshold, it indicates that the entity data in the entity dataset is a low-quality mention. If the quality assessment result is greater than the quality assessment threshold, it indicates that the entity data in the entity dataset is a high-quality mention. Then, the entity dataset can be used to annotate the second text dataset to be labeled in the text dataset to obtain the labeled dataset. The second text dataset can be a set of entity data whose quality assessment result is greater than the quality assessment threshold, and the labeled dataset can be pseudo-labeled data.

[0037] For example, if the quality assessment threshold is set to 0.5, and the quality assessment result of entity data in the entity dataset is 0.2, which is less than the quality assessment threshold of 0.5, then the entity data is considered a low-quality mention; if the quality assessment result of entity data in the entity dataset is 0.6, which is greater than the quality assessment threshold of 0.5, then the entity data is considered a high-quality mention.

[0038] Optionally, when labeling the second text dataset to be labeled in the text dataset using the entity dataset, a matching method can be used to match each entity data in the entity dataset with the second text dataset to obtain the labeled dataset. The labeled dataset may include the matched entity data in the second text dataset and the position of the matched entity data in the second text dataset.

[0039] For example, the second text dataset to be labeled is: [{"text": "text to be mined"}, {"text": "text to be mined"}], the set of mined mentions is: ["XX", "biotechnology company"], and the labeled dataset after labeling is: [{"text": "XX is a biotechnology company", "set of mined mentions": [{"content of mention": "XX", "position of mention in the original text": [0,4]}, {"content of mention": "biotechnology company", "position of mention in the original text": [2,8]}]}].

[0040] It should be noted that the above content is only for illustrative purposes, and the annotation method of the second text dataset is not limited here. Any method and process used to annotate the second text dataset is within the protection scope of this disclosure, and will not be elaborated here.

[0041] In this embodiment, the training data for the quality assessment model can be open-source named entity recognition data. This training data can be divided into positive example data and negative example data. Positive example data refers to mentions with accurate boundaries, while negative example data refers to mentions with inaccurate boundaries. Inaccurate mentions can be constructed by expanding or shortening them to the left or right using mentions with accurate boundaries.

[0042] Optionally, this embodiment can select a single-tower model or a dual-tower model as the model framework for the aforementioned mention quality assessment model. Then, the text corresponding to the positive example data or the text corresponding to the negative example data is concatenated to form new text. Finally, binary classification training is performed on the positive example data or the negative example data to generate the mention quality assessment model. It should be noted that the training method of the mention quality assessment model is not limited here. Any method and process used to train the mention quality assessment model is within the protection scope of this disclosure, and will not be elaborated here.

[0043] Step S108: Output the labeled dataset.

[0044] In the technical solution provided in step S108 of this disclosure, after labeling the second text dataset to be labeled in the text dataset using the entity dataset to obtain the labeled dataset, the labeled dataset can be verified, and the successfully verified labeled dataset can be output. This successfully verified labeled dataset can be used to obtain an entity recognition model through supervised training. Since the entity recognition model can be used to automatically identify entity data in the text to be recognized, the technical effect of improving the development efficiency of text data and reducing the development cost of text data is achieved.

[0045] Through steps S102 to S108, the first text dataset to be converted is obtained from the text dataset; the first text dataset is converted into an entity dataset; in response to the quality assessment result of the entity data in the entity dataset being greater than the quality assessment threshold, the second text dataset to be labeled in the text dataset is labeled using the entity dataset to obtain a labeled dataset; the labeled dataset is then output. In other words, an entity dataset is first mined from the first text dataset, and then the mined entity dataset is used to label the second text dataset to obtain a labeled dataset. This labeled dataset can be used to obtain a mention recognition model through supervised training to achieve the goal of obtaining high-quality mentions, thereby solving the technical problem of low development efficiency of text data and achieving the technical effect of improving the development efficiency of text data.

[0046] The method described in this embodiment will now be described in further detail.

[0047] As an optional implementation, step S104, converting the first text dataset into an entity dataset, includes: determining the entity dataset based on candidate entity data in the first text dataset whose frequency of occurrence is higher than a first frequency threshold.

[0048] In this embodiment, a first frequency threshold can be set to filter candidate entity data that appears frequently in the first text dataset. The set of multiple candidate entity data that appear more frequently than the first frequency threshold in the first text dataset is determined as the entity dataset. This achieves the purpose of filtering high-frequency mentions in the first text dataset, thereby improving the technical effect of mention mining quality. The candidate entity data can be high-frequency mention data, such as high-frequency phrases.

[0049] For example, if the word "sun" appears 5 times and "sunlight" appears 3 times in a text, and the first frequency threshold is set to 4, then the frequency of "sun" in the text is greater than the first frequency threshold. Therefore, "sun" in the text can be considered as candidate entity data with a frequency higher than the first frequency threshold. In other words, "sun" can be added to the entity dataset.

[0050] Optionally, the entity data with higher frequency can be relatively common or well-known words or phrases, such as "window", while the entity data with lower frequency can be uncommon or relatively obscure words or phrases, such as "pavilion".

[0051] It should be noted that the above method for determining the entity dataset from the first text dataset is only an example, and there is no limitation on the method for determining the entity dataset from the first text dataset. Any method and process used to determine the entity dataset from the first text dataset is within the protection scope of this disclosure, and will not be described in detail here.

[0052] As an optional implementation, determining an entity dataset based on candidate entity data in a first text dataset that appears more frequently than a first frequency threshold includes: obtaining target candidate entity data including candidate entity data from the first text dataset; and determining the entity dataset based on the target candidate entity data in the first text dataset that appears more frequently than a second frequency threshold.

[0053] In this embodiment, the second frequency threshold is less than the first frequency threshold. After filtering candidate entity data whose frequency of occurrence is higher than the first frequency threshold, target candidate entity data including candidate entity data can be determined from the first text dataset. Then, the second frequency threshold is set, and the target candidate entity data in the first text dataset is filtered by the second frequency threshold. The set of multiple target candidate entity data in the first text dataset whose frequency of occurrence is higher than the second frequency threshold is determined as the entity dataset. This achieves the purpose of secondary verification of high-frequency mentions in the first text dataset, thereby realizing the technical effect of improving the mention mining quality.

[0054] Optionally, the window length of the target candidate entity data is greater than the window length of the candidate entity data. For example, the target candidate entity data can be "XX University" which appears 4 times in the text, and the candidate entity data can be "XX" which appears 10 times in the text. In order to improve the quality of mention mining, we can first use the first frequency threshold to mine high-frequency and short-term seeds, and then use the second frequency threshold to perform secondary hit verification. For example, we can first set the first frequency threshold to 8 to filter out "XX" in the text, and then set the second frequency threshold to 4 to filter out "XX University" in the text and add it to the entity dataset.

[0055] It should be noted that the above method is an unsupervised method and does not require supervised corpus. For example, the input is: [{"text": "text to be mined"}, {"text": "text to be mined"}], and the output is: [{"text": "text to be mined", "mined mention set": [{"mention content": "test", "mention position in the original text": [1,3]}, [{"text": "text to be mined", "mined mention set": [{"mention content": "test", "mention position in the original text": [2,5]}]}].

[0056] It should be noted that the above method for determining the entity dataset based on candidate entity data that appears more frequently than the first frequency threshold in the first text dataset is only an example. This method does not limit the determination of the entity dataset based on candidate entity data that appears more frequently than the first frequency threshold in the first text dataset. Any method and process used to determine the entity dataset based on candidate entity data that appears more frequently than the first frequency threshold in the first text dataset is within the protection scope of this disclosure, and will not be elaborated here.

[0057] As an optional implementation, step S104, converting the first text dataset into an entity dataset, includes: performing word segmentation on the first text dataset to obtain a word sequence; and determining the entity dataset based on the words in the word sequence that match the target part of speech.

[0058] In this embodiment, the target part of speech can be the pattern of the part of speech of the segmented words in the word sequence. The part of speech of the segmented words can be formed by combining the symbol "-", for example, noun-noun (Noun-Noun, abbreviated as nn). The part of speech of the segmented words can include nouns, verbs, numerals, etc. This is only an example and is not specifically limited.

[0059] In this embodiment, a part-of-speech tagging tool can be used to segment the first text dataset to obtain a word sequence. Then, based on the words in the word sequence that match the target part of speech, the entity dataset is determined to achieve the purpose of mention mining based on the word sequence, thereby achieving the technical effect of improving the quality of mention mining.

[0060] For example, the word sequence obtained by segmenting "XX is a technology company" can be "XX / is / a / technology / company".

[0061] For example, if the target part of speech is "noun-noun", and the first text in the first text dataset contains "technology company" (part of speech "noun-noun"), and the second text contains "XX university" (part of speech "noun-noun"), then the words in the first text dataset that match the target part of speech are "technology company" and "XX university". In other words, "technology company" and "XX university" can be identified as entity data in the entity dataset.

[0062] It should be noted that the above method of converting the first text dataset into an entity dataset through word segmentation is only an example. There is no limitation on the method of converting the first text dataset into an entity dataset through word segmentation. Any method and process of converting the first text dataset into an entity dataset through word segmentation is within the protection scope of this disclosure, and will not be described in detail here.

[0063] As an optional implementation, the method may further include: acquiring part-of-speech tagging data; and identifying target part-of-speech combinations in the part-of-speech tagging data that appear more frequently than a third frequency threshold as target parts of speech.

[0064] In this embodiment, after performing part-of-speech tagging on the first text dataset, part-of-speech tagging data can be obtained. Then, a third frequency threshold is set, and part-of-speech combinations with high frequency in the part-of-speech tagging data are filtered through the third frequency threshold. Target part-of-speech combinations with a frequency greater than the third frequency threshold are identified as target parts of speech, thereby achieving the purpose of mention mining based on target parts of speech. This achieves the technical effect of improving the quality of mention mining. The part-of-speech tagging data can include entity data tagged with part-of-speech identifiers. Part-of-speech identifiers can be used to represent the part of speech of entity data, such as nouns, verbs, numerals, etc. The target part of speech can be the mined part-of-speech template.

[0065] For example, the part-of-speech tagging data obtained after tagging "XX is a technology company" can be "XX (noun) / is (verb) / a (numeral) / technology (noun) / company (noun)".

[0066] For another example, if the third frequency threshold is set to 2, and the first text in the first text dataset contains "technology company" with a part of speech of "noun-noun", the second text contains "XX university" with a part of speech of "noun-noun", and the third text contains "XX primary school" with a part of speech of "noun-noun", that is, the frequency of "noun-noun" in the first text dataset is 3. The frequency of "noun-noun" is greater than the third frequency threshold, so "noun-noun" can be identified as the target part of speech.

[0067] It should be noted that the above methods for determining the target part of speech are merely illustrative examples, and no limitation is made here. Any methods and processes used to determine the target part of speech are within the scope of protection of this disclosure, and will not be elaborated here.

[0068] As an optional implementation, determining the entity dataset based on words in the word sequence that match the target part of speech includes: combining words in the word sequence that match different target parts of speech to obtain the entity dataset.

[0069] In this embodiment, words that match different target parts of speech in a word sequence can be combined to obtain entity datasets with different target parts of speech, thereby achieving the technical effect of improving the quality of mention mining.

[0070] For example, by combining words in a word sequence that match the target part of speech "noun-noun", we can mine entity datasets that match "noun-noun"; by combining words in a word sequence that match the target part of speech "noun-verb", we can mine entity datasets that match "noun-verb".

[0071] It should be noted that the above method for combining words in a word sequence that match different target parts of speech is only an example. This method does not limit the scope of this disclosure. Any method and process for combining words in a word sequence that match different target parts of speech is within the scope of protection of this disclosure.

[0072] As an optional implementation, step S104, converting the first text dataset into an entity dataset, includes: performing word segmentation on the first text dataset to obtain a word sequence; and determining the entity dataset based on the words in the word sequence that match the dependency grammar relations.

[0073] In this embodiment, dependency grammar relations can be represented as a pattern of grammar relations between word segments in a word sequence. The grammar relations between word segments can be assembled by the symbol "-", for example, attribute-verb object (ATT-VOB). The grammar relations between word segments can include subject-predicate relations, attribute-verb relations, verb-object relations, core relations, and indirect object relations, etc. This is only an example and is not specifically limited.

[0074] In this embodiment, a dependency parsing tool can be used to perform word segmentation and dependency parsing on the first text dataset to obtain a word sequence. Then, based on the words in the word sequence that match the dependency grammatical relations, the entity dataset is determined to achieve the purpose of mention mining based on dependency grammatical relations, thereby achieving the technical effect of improving the quality of mention mining.

[0075] For example, if the dependency grammar relation is "attributive-verb-object relation", and the first text in the first text dataset contains "technology company", with the grammar relation being "attributive-verb-object relation", and the second text contains "XX university", with the grammar relation being "attributive-verb-object relation", then the words in the first text dataset that match the dependency grammar relation are "technology company" and "XX university". In other words, "technology company" and "XX university" can be identified as entity data in the entity dataset.

[0076] It should be noted that the above method for converting the first text dataset into an entity dataset based on word segmentation and dependency analysis is only an example. This method is not limited to the method of converting the first text dataset into an entity dataset based on word segmentation and dependency analysis. Any method and process for converting the first text dataset into an entity dataset based on word segmentation and dependency analysis is within the protection scope of this disclosure, and will not be described in detail here.

[0077] As an optional implementation, the method may further include: acquiring relation annotation data; and combining dependency syntax relations that appear more frequently than a fourth frequency threshold in the relation annotation data to determine them as dependency syntax relations.

[0078] In this embodiment, after performing word segmentation and dependency parsing on the first text dataset, relation annotation data can be obtained. Then, a fourth frequency threshold is set, and the dependency grammar relation combinations that appear frequently in the relation annotation data are filtered through the fourth frequency threshold. Dependency grammar relation combinations that appear more frequently than the fourth frequency threshold are identified as dependency grammar relations, so as to achieve the purpose of mention mining based on dependency grammar relations, thereby achieving the technical effect of improving the quality of mention mining. The relation annotation data may include entity data annotated by dependency grammar relation identifiers. Dependency grammar relation identifiers can be used to represent the grammatical relations of entity data, such as subject-verb relations, attributive-head relations, verb-object relations, core relations, and indirect object relations. The dependency grammar relations can be the templates of the mined dependency grammar relations.

[0079] For example, the relational tagging data obtained after segmenting and dependency parsing "XX is a technology company" can be "XX (subject-verb relationship) / is (core relationship) / a (attributive-head relationship) / technology (attributive-head relationship) / company (verb-object relationship)".

[0080] For another example, if the fourth frequency threshold is set to 2, and the first text in the first text dataset contains "technology company", the dependency grammar relationship is "attributive-verb-object relationship", the second text contains "XX university", the dependency grammar relationship is "attributive-verb-object relationship", and the third text contains "XX primary school", the dependency grammar relationship is "attributive-verb-object relationship". That is, the frequency of "attributive-verb-object relationship" in the first text dataset is 3. The frequency of "attributive-verb-object relationship" is greater than the fourth frequency threshold, so "attributive-verb-object relationship" can be identified as a dependency grammar relationship.

[0081] It should be noted that the above methods for determining dependency syntactic relations are merely illustrative examples, and no limitation is made on the methods for determining dependency syntactic relations here. Any methods and processes used to determine dependency syntactic relations are within the protection scope of this disclosure, and will not be elaborated here.

[0082] As an optional implementation, determining the entity dataset based on words in the word sequence that match dependency grammar relations includes: combining words in the word sequence that match different dependency grammar relations to obtain the entity dataset.

[0083] In this embodiment, words that match different dependency grammar relations in a word sequence can be combined to obtain entity datasets with different dependency grammar relations, thereby achieving the technical effect of improving the quality of mention mining.

[0084] For example, by combining words in a word sequence that match the dependency grammar relation "subject-predicate relation-core relation", entity datasets matching the "subject-predicate relation-core relation" can be mined; similarly, by combining words in a word sequence that match the dependency grammar relation "attributive-head relation-verb-object relation", entity datasets matching the "attributive-head relation-verb-object relation" can be mined.

[0085] It should be noted that the above method of combining words in a word sequence that match different dependency grammatical relations is only an example. This method does not limit the scope of the method of combining words in a word sequence that match different dependency grammatical relations. Any method and process for combining words in a word sequence that match different dependency grammatical relations is within the scope of protection of this disclosure, and will not be elaborated here.

[0086] As an optional implementation, step S106, which uses the entity dataset to annotate the second text dataset to obtain the annotated dataset, includes: matching the second text dataset with each entity data in the entity dataset to obtain the annotated dataset.

[0087] In this embodiment, the second text dataset can be matched using each entity data in the entity dataset to obtain a labeled dataset, thereby achieving the technical effect of labeling the second text dataset with the mined mentions. The labeled dataset can be a pseudo-labeled dataset, and can include the entity data matched in the second text dataset and the position of the matched entity data in the second text dataset.

[0088] For example, the second text dataset to be labeled is: [{"Text": "Text to be mined"}, {"Text": "Text to be mined"}], the set of mined mentions is: ["XX", "Technology Company"], and the labeled dataset is: [{"Text": "XX is a technology company", "Minement mention set (Phrase List)": [{"Content of mention (Value): "XX", "Position of mention in the original text (Offset)": [0,2]}, {"Content of mention": "Technology Company", "Position of mention in the original text": [5,9]}]}].

[0089] It should be noted that the annotation method of the second text dataset mentioned above is only an example, and the annotation method of the second text dataset is not limited here. Any method and process used to annotate the second text dataset is within the protection scope of this disclosure, and will not be described in detail here.

[0090] As an optional implementation, the method may further include: verifying the labeled dataset in response to a verification operation instruction for the labeled dataset; and outputting the labeled dataset including: outputting a labeled dataset that has been successfully verified.

[0091] In this embodiment, to avoid erroneous labeled data in the labeled dataset, a crowdsourcing verification method can be adopted. The labeled dataset is verified by a verification operation command, and the successfully verified labeled dataset is output to improve the technical effect of labeling quality. The verification operation command can be used to verify the quality of the labeled data in the labeled dataset.

[0092] The method for determining the model according to the embodiments of this disclosure will be described below.

[0093] Figure 2(a) is a flowchart of a method for determining a model according to an embodiment of the present disclosure. As shown in Figure 2(a), the method may include the following steps:

[0094] Step S202: Obtain the labeled dataset.

[0095] In the technical solution provided by step S202 of this disclosure, an labeled dataset can be obtained. The labeled dataset can be a dataset obtained by labeling the text dataset to be labeled in the text dataset using the entity dataset. The labeled dataset can include the entity dataset. The quality assessment result of the entity data in the entity dataset is greater than the quality assessment threshold. The entity dataset can be obtained by converting the text dataset to be converted in the text dataset.

[0096] Step S204: Supervised training is performed based on the labeled dataset to obtain the entity recognition model.

[0097] In the technical solution provided by step S204 of this disclosure, supervised training can be performed using labeled datasets to obtain an entity recognition model. Then, the entity recognition model can identify entity data in the text to be recognized, thereby achieving the technical effect of improving the development efficiency of text data and reducing the development cost of text data.

[0098] Figure 2(b) is a schematic diagram of a computer product for performing a method for determining a model according to an embodiment of the present disclosure. As shown in Figure 2(b), the above steps S202 to S204 can be performed by a computing device, the labeled dataset is input into the computing device, the computing device performs supervised training based on the labeled dataset, and outputs an entity recognition model, so as to achieve the technical effect of improving the development efficiency of text data and reducing the development cost of text data.

[0099] Figure 2(c) is a schematic diagram of a human-computer interaction interface for performing a model determination method according to an embodiment of the present disclosure. As shown in Figure 2(c), the labeled dataset 202 can be input by triggering the control 201 on the interface, and the entity recognition model 203 output after supervised training of the labeled dataset can be obtained.

[0100] It should be noted that no specific limitations are made here on the computer products and human-computer interaction interfaces used to perform the method of determining the model. Any computer products and human-computer interaction interfaces used to perform the method of determining the model are within the protection scope of this disclosure, and will not be listed here.

[0101] Through steps S202 to S204 above, an labeled dataset is obtained. The labeled dataset is obtained by labeling the text dataset to be labeled in the text dataset using the entity dataset, and includes the entity dataset. The quality assessment result of the entity data in the entity dataset is greater than the quality assessment threshold. The entity dataset is converted from the text dataset to be converted in the text dataset. Supervised training is performed based on the labeled dataset to obtain an entity recognition model. The entity recognition model is used to identify entity data in the text to be recognized, thereby solving the technical problem of low development efficiency of text data and achieving the technical effect of improving the development efficiency of text data.

[0102] The text data processing method of this disclosure will be further described below with reference to preferred embodiments.

[0103] As enterprises become increasingly digitalized, their demand for internal data retrieval and analysis is also growing. This includes internal documents and data, such as purchase orders, industry reports, business attachments, sales contracts, employment agreements, commercial invoices, and resumes.

[0104] Currently, extracting high-quality mentions plays a crucial role in tasks such as information extraction, retrieval, classification modeling, and topic analysis. Therefore, obtaining high-quality mentions is of great significance for enterprise-level applications.

[0105] In related technologies, a common method for extracting high-quality mentions is sequence labeling technology, such as Named Entity Recognition (NER). However, this technology mainly trains the mention recognition model through labeled data, which requires a large amount of high-quality labeled data. But when enterprise customers apply it, it is usually in vertical scenarios, where labeled data is lacking, and the cost of data labeling is very high.

[0106] However, embodiments of this disclosure provide a knowledge-enhanced semi-supervised mention mining technique. Figure 3 This is a schematic diagram of a semi-supervised mention mining system according to an embodiment of the present disclosure, such as... Figure 3 As shown, the system may include: a data (document) set 301, an unsupervised mining module 302, a semi-supervised mining module 303, a supervised training module 304, and a mention module 305.

[0107] Data (document) set 301 is mainly a collection of data to be mined, where data refers to text or documents.

[0108] The unsupervised mining module 302 includes: high-frequency co-occurrence phrase mining 3021, used to filter high-frequency phrases co-occurring in multiple sentences; and mention quality assessment model 3022, used to score mention quality and filter out low-scoring mentions, which can further improve mention quality.

[0109] The semi-supervised mining module 303 includes: part-of-speech template mining 3031 and dependency relation template mining 3032, which are used to mine mentions based on a small amount of labeled data and generate candidate mentions; and mention quality assessment model 3033, which is used to score mention quality and filter out low-scoring mentions, which can further improve mention quality.

[0110] The supervised training module 304 includes: a labeled dataset 3041, which can generate a labeled dataset for mentions mined using unsupervised or semi-supervised mining methods through matching, wherein the labeled dataset can be pseudo-labeled data; a crowdsourcing (judgment) confirmation 3042, which can quickly make judgments by finding annotators through crowdsourcing when it is necessary to improve the quality of unlabeled data; and a sequence labeling model 3043, which is used as a mention recognition model after the data verification is completed.

[0111] The following section provides a further introduction to the knowledge-enhanced semi-supervised mention mining method.

[0112] High-frequency co-occurrence phrase mining is a commonly used method in unsupervised phrase mining. First, candidate phrases are generated from the input text using the language model (N-Gram) method. Then, the frequency of candidate phrases in each text is counted. Finally, a relatively high threshold is set for filtering. The window length of the language model can be adjusted according to the characteristics of the data, for example, greater than 3 and less than 15.

[0113] It should be noted that the above-mentioned high-frequency co-occurrence phrase mining method can mine high-frequency mentions in the text set. However, when the set threshold is low, the quality of the mined mentions is relatively low. Therefore, it is advisable to first mine seeds of high-frequency short language models (e.g., window length of 3-4) and then lower the threshold. However, when lowering the threshold, it is necessary to perform secondary hit verification in conjunction with high-frequency seeds.

[0114] In this embodiment of the disclosure, a part-of-speech tagging tool can be used to segment and tag the text to be mined. For example, input: XX is a technology company, output: XX (noun) / is (verb) / a (numeral) / technology (noun) / company (noun).

[0115] Assuming that “technology company” is a phrase in the above input, we can form a pattern by combining the parts of speech of the known phrase in the sentence. For example, the pattern of “technology company” is “noun-noun”, that is, the parts of speech of the corresponding words are combined by the symbol “-”.

[0116] By performing the above operations on a small amount of known labeled data, a frequency statistic of a pattern is generated. The pattern set obtained by filtering the high-frequency patterns through a threshold is the pattern template mined.

[0117] For unlabeled data, the above part-of-speech tagging method can be used to segment words, and then the mined patterns can be matched on the segmented sequences. The word sequences mined from each pattern are combined together to form the mined words.

[0118] In this embodiment of the disclosure, a dependency parsing tool can also be used to perform word segmentation and dependency analysis on the text to be mined. For example: Input: XX is a technology company, Output: XX (subject-verb relationship) / is (core relationship) / a (attributive-head relationship) / technology (attributive-head relationship) / company (verb-object relationship).

[0119] Assuming that “technology company” is a phrase in the above input, the dependency relations of the word segments that the known phrase appears in the sentence can be combined into a pattern. For example, the pattern of “technology company” is “attributive-head relation-verb-object relation”, that is, the dependency relations after the corresponding word segmentation are combined by the symbol “-”.

[0120] By performing the above operations on a small amount of known labeled data, a frequency statistic of a pattern is generated. The pattern set obtained by filtering the high-frequency patterns through a threshold is the pattern template mined.

[0121] For unlabeled data, the above part-of-speech tagging method can be used for dependency analysis. Then, the mined pattern matching is performed on the segmented sequence. The word sequences mined from each pattern are combined together to form the mined words.

[0122] In this embodiment of the disclosure, the accuracy of the mention boundaries obtained by unsupervised mention mining or template mining is the main reason for the low quality of the mining results. Therefore, a mention quality assessment model that evaluates from the perspective of mention boundaries can be trained. The training data of the mention quality assessment model can be constructed using open-source named entity recognition data. The constructed data is divided into two types: positive example data and negative example data. Positive example data can be mentions with accurate boundaries, and negative example data can be mentions with inaccurate boundaries. Such mentions can be constructed by expanding or shortening the accurate mention boundaries to the left or right.

[0123] Optionally, positive or negative mentions can be combined with their corresponding text to form new text, and then binary classification training can be performed on the training data.

[0124] Figure 4 This is a schematic diagram of a model framework for a quality assessment model according to an embodiment of this disclosure, such as... Figure 4 As shown, either the single-tower model 401 or the dual-tower model 402 can be used for mention quality assessment. During prediction, the model can output the score obtained through similarity calculation, and then set a threshold. If the score is greater than the threshold, it is a high-quality mention; otherwise, it is a low-quality mention.

[0125] In this embodiment of the disclosure, the mined mentions can be labeled by matching to obtain pseudo-labeled data. For example, the input is: data to be labeled: [{"text": "text to be mined"}, {"text": "text to be mined"}], the set of mined mentions: ["XX", "technology company"], and the output is: [{"text": "XX is a technology company", "set of mined mentions": [{"content of mention": "XX", "position of mention in the original text": [0,2]}, {"content of mention": "technology company", "position of mention in the original text": [5,9]}]}].

[0126] It should be noted that the data labeled using the above method often contains erroneous labels, hence the term pseudo-labeled data. In other words, the labeling results are not entirely accurate and may generally require manual verification in order to improve the technical effect of labeling data quality.

[0127] In this embodiment of the disclosure, pseudo-labeled data can also be generated through automatic data annotation. Since pseudo-labeled data is prone to errors and has a significant impact on the accuracy of trained supervised models, the quality of pseudo-labeled data can be verified manually. Compared with direct annotation, verifying the quality of pseudo-labeled data is more efficient.

[0128] Alternatively, supervised model training can use currently popular sequence labeling models, and the base can be a pre-trained model (Ernie), which has a good knowledge enhancement effect.

[0129] Optionally, unsupervised mining methods refer to mention mining using a combination of "high-frequency co-occurrence phrase mining" and "phrase quality assessment model"; semi-supervised mining methods refer to mention mining using a combination of "part-of-speech template mining" and "dependency relation template mining" and "phrase quality assessment model".

[0130] Optionally, a certain amount of pseudo-labeled data can be generated by using unsupervised or supervised methods to mine mentions. After crowdsourcing confirmation, a supervised mention model can be trained.

[0131] In this embodiment, text is first acquired, and mention mining is performed on the acquired text using an unsupervised mining method or a semi-supervised mining method. The mined mentions are then labeled on the text using automatic data annotation to generate pseudo-labeled data. The quality of the generated pseudo-labeled data is then confirmed by manual verification. Finally, supervised model training is performed using the crowdsourced confirmed pseudo-labeled data to achieve the goal of obtaining high-quality mentions. This solves the technical problem of low development efficiency of text data and achieves the technical effect of improving the development efficiency of text data.

[0132] This disclosure also provides an embodiment for performing Figure 1 The text data processing method apparatus of the embodiment shown.

[0133] Figure 5 This is a schematic diagram of a text data processing apparatus according to an embodiment of the present disclosure, such as... Figure 4 As shown, the text data processing device 500 may include: a first acquisition unit 501, a conversion unit 502, a labeling unit 503, and an output unit 504.

[0134] The first acquisition unit 501 is used to acquire the first text dataset to be converted from the text dataset.

[0135] The conversion unit 502 is used to convert the first text dataset into an entity dataset.

[0136] The annotation unit 503 is used to annotate the second text dataset to be annotated in the text dataset in response to the quality assessment result of the entity data in the entity dataset being greater than the quality assessment threshold, thereby obtaining the annotation dataset, wherein the annotation dataset includes the entity dataset.

[0137] Output unit 504 is used to output a labeled dataset, wherein the labeled dataset is used to obtain an entity recognition model through supervised training, and the entity recognition model is used to identify entity data in the text to be recognized.

[0138] Optionally, the conversion unit 502 includes: a first determining module, used to determine an entity dataset based on candidate entity data in the first text dataset whose frequency of occurrence is higher than a first frequency threshold.

[0139] Optionally, the first determining module includes: an acquisition submodule, used to acquire target candidate entity data including candidate entity data from the first text dataset; and a determining submodule, used to determine an entity dataset based on the target candidate entity data in the first text dataset whose frequency of occurrence is higher than a second frequency threshold, wherein the second frequency threshold is less than the first frequency threshold.

[0140] Optionally, the conversion unit 502 includes: a first processing module for performing word segmentation on the first text dataset to obtain a word sequence; and a second determining module for determining the entity dataset based on the words in the word sequence that match the target part of speech.

[0141] Optionally, the conversion unit 502 includes: a first acquisition module, used to acquire part-of-speech tagging data, wherein the part-of-speech tagging data includes entity data tagged by part-of-speech identifiers; and a third determination module, used to determine the target part-of-speech by identifying the target part-of-speech combination in the part-of-speech tagging data whose frequency of occurrence is greater than a third frequency threshold.

[0142] Optionally, the second determining module includes: a first combining module, used to combine words in the word sequence that match different target parts of speech to obtain an entity dataset.

[0143] Optionally, the conversion unit 502 includes: a second processing module for performing word segmentation on the first text dataset to obtain a word sequence; and a fourth determination module for determining the entity dataset based on the words in the word sequence that match the dependency grammar relations.

[0144] Optionally, the conversion unit 502 includes: a second acquisition module, used to acquire relation annotation data, wherein the relation annotation data includes entity data annotated by dependency syntax relation identifiers; and a fifth determination module, used to determine the combination of dependency syntax relations in the relation annotation data that occur more frequently than a fourth frequency threshold as dependency syntax relations.

[0145] Optionally, the fourth determining module includes: a second combining module, used to combine words in the word sequence that match different dependency grammatical relations to obtain an entity dataset.

[0146] Optionally, the annotation unit 503 includes: a matching module, used to match the second text dataset with each entity data in the entity dataset to obtain an annotation dataset, wherein the annotation dataset includes the entity data matched in the second text dataset and the position of the matched entity data in the second text dataset.

[0147] Optionally, the apparatus further includes a verification unit for verifying the labeled dataset in response to a verification operation command for the labeled dataset.

[0148] Optionally, the output unit 504 includes an output module for outputting the labeled dataset that has been successfully verified.

[0149] In the text data processing apparatus of this disclosure embodiment, a first acquisition unit is used to acquire a first text dataset to be converted from the text dataset; a conversion unit is used to convert the first text dataset into an entity dataset; a labeling unit is used to label a second text dataset to be labeled from the text dataset using the entity dataset in response to the entity dataset having a quality assessment result greater than a quality assessment threshold, thereby obtaining a labeled dataset, wherein the labeled dataset includes the entity dataset; and an output unit is used to output the labeled dataset, wherein the labeled dataset is used to obtain an entity recognition model through supervised training, and the entity recognition model is used to recognize entity data in the text to be recognized, thereby solving the technical problem of low development efficiency of text data and achieving the technical effect of improving the development efficiency of text data.

[0150] This disclosure also provides a model determination apparatus for performing the model determination method of the embodiment shown in FIG2.

[0151] Figure 6 This is a schematic diagram of a model determining device according to an embodiment of the present disclosure, such as... Figure 6 As shown, the model determination device 600 may include a second acquisition unit 601 and a training unit 602.

[0152] The second acquisition unit 601 is used to acquire the labeled dataset, wherein the labeled dataset is obtained by labeling the text dataset to be labeled in the text dataset using the entity dataset, and includes the entity dataset, wherein the quality assessment result of the entity data in the entity dataset is greater than the quality assessment threshold, and the entity dataset is obtained by converting the text dataset to be converted in the text dataset.

[0153] Training unit 602 is used for supervised training based on labeled datasets to obtain entity recognition model, wherein the entity recognition model is used to identify entity data in the text to be recognized.

[0154] In the text data processing apparatus of this embodiment, the second acquisition unit is used to acquire a labeled dataset, wherein the labeled dataset is obtained by labeling the text dataset to be labeled in the text dataset using an entity dataset, and includes an entity dataset, wherein the quality assessment result of the entity data in the entity dataset is greater than the quality assessment threshold, and the entity dataset is obtained by converting the text dataset to be converted in the text dataset; the training unit is used to perform supervised training based on the labeled dataset to obtain an entity recognition model, wherein the entity recognition model is used to recognize entity data in the text to be recognized, thereby solving the technical problem of low development efficiency of text data and achieving the technical effect of improving the development efficiency of text data.

[0155] The acquisition, storage, and application of user personal information involved in the technical solution disclosed herein comply with the provisions of relevant laws and regulations and do not violate public order and good morals.

[0156] According to embodiments of this disclosure, this disclosure also provides an electronic device, a readable storage medium, and a computer program product.

[0157] Figure 7 This is a schematic diagram of an electronic device for a text data processing method according to an embodiment of the present disclosure. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device may also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the present disclosure described and / or claimed herein.

[0158] like Figure 7 As shown, device 700 includes a computing unit 701, which can perform various appropriate actions and processes based on a computer program stored in read-only memory (ROM) 702 or a computer program loaded from storage unit 708 into random access memory (RAM) 703. RAM 703 may also store various programs and data required for the operation of device 700. The computing unit 701, ROM 702, and RAM 703 are interconnected via bus 704. Input / output (I / O) interface 705 is also connected to bus 704.

[0159] Multiple components in device 700 are connected to I / O interface 705, including: input unit 706, such as keyboard, mouse, etc.; output unit 707, such as various types of monitors, speakers, etc.; storage unit 708, such as disk, optical disk, etc.; and communication unit 709, such as network card, modem, wireless transceiver, etc. Communication unit 709 allows device 700 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0160] The computing unit 701 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the computing unit 701 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. The computing unit 701 performs the various methods and processes described above, such as text data processing methods. For example, in some embodiments, the text data processing methods may be implemented as computer software programs tangibly contained in a machine-readable medium, such as storage unit 708. In some embodiments, part or all of the computer program may be loaded and / or installed on device 700 via ROM 702 and / or communication unit 709. When the computer program is loaded into RAM 703 and executed by the computing unit 701, one or more steps of the text data processing methods described above may be performed. Alternatively, in other embodiments, the computing unit 701 may be configured to perform text data processing methods by any other suitable means (e.g., by means of firmware).

[0161] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-a-chip (SoCs), payload-programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0162] The program code used to implement the methods of this disclosure may be written in any combination of one or more programming languages. This program code may be provided to a processor or controller of a general-purpose computer, special-purpose computer, or other programmable data processing apparatus, such that when executed by the processor or controller, the program code causes the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The program code may be executed entirely on a machine, partially on a machine, as a standalone software package partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0163] In the context of this disclosure, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can be, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0164] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device for displaying information to the user (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor); and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the computer. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0165] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as a data server), or computing systems that include middleware components (e.g., an application server), or computing systems that include frontend components (e.g., a user computer with a graphical user interface or web browser through which a user can interact with embodiments of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., a communication network). Examples of communication networks include local area networks (LANs), wide area networks (WANs), and the Internet.

[0166] Computer systems can include clients and servers. Clients and servers are generally located far apart and typically interact via communication networks. Client-server relationships are created by computer programs running on the respective computers and having a client-server relationship with each other. Servers can be cloud servers, servers in distributed systems, or servers incorporating blockchain technology.

[0167] It should be understood that the various forms of processes shown above can be used to rearrange, add, or delete steps. For example, the steps described in this disclosure can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution disclosed in this disclosure can be achieved, and this is not limited herein.

[0168] The specific embodiments described above do not constitute a limitation on the scope of protection of this disclosure. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this disclosure should be included within the scope of protection of this disclosure.

Claims

1. A method for processing text data, comprising: Retrieve the first text dataset to be converted from the text dataset; The entity dataset is determined based on candidate entity data in the first text dataset that appear more frequently than a first frequency threshold; In response to the entity data in the entity dataset having a quality assessment result greater than the quality assessment threshold, the entity dataset is used to annotate the second text dataset to be labeled in the text dataset to obtain an annotated dataset, wherein the annotated dataset includes the entity dataset; Output the labeled dataset, wherein the labeled dataset is used to obtain an entity recognition model through supervised training, and the entity recognition model is used to identify entity data in the text to be recognized; The method of determining an entity dataset based on candidate entity data in the first text dataset that appears more frequently than a first frequency threshold includes: obtaining target candidate entity data including the candidate entity data from the first text dataset; and determining the entity dataset based on the target candidate entity data in the first text dataset that appears more frequently than a second frequency threshold, wherein the second frequency threshold is less than the first frequency threshold.

2. The method according to claim 1, wherein, The method further includes: converting the first text dataset into an entity dataset, including: The first text dataset is segmented into words to obtain word sequences; The entity dataset is determined based on the words in the word sequence that match the target part of speech.

3. The method according to claim 2, further comprising: Obtain part-of-speech tagging data, wherein the part-of-speech tagging data includes entity data tagged with part-of-speech identifiers; The target part-of-speech tagging data is determined by the combination of target part-of-speech tags that appears more frequently than the third frequency threshold.

4. The method according to claim 2, wherein, Based on the words in the word sequence that match the target part of speech, the entity dataset is determined to include: The entity dataset is obtained by combining words in the word sequence that match different target parts of speech.

5. The method according to claim 1, wherein converting the first text dataset into an entity dataset comprises: The first text dataset is segmented into words to obtain word sequences; The entity dataset is determined based on the words in the word sequence that match the dependency grammar relations.

6. The method according to claim 5, further comprising: Obtain relation annotation data, wherein the relation annotation data includes entity data annotated using dependency syntax relation identifiers; The dependency syntax relations that appear more frequently than the fourth frequency threshold in the relation annotation data are combined and identified as the dependency syntax relations.

7. The method according to claim 5, wherein, Based on the words in the word sequence that match dependency grammar relations, the entity dataset is determined to include: The entity dataset is obtained by combining words in the word sequence that match different dependency grammar relations.

8. The method according to claim 1, wherein, The entity dataset is used to annotate the second text dataset to be annotated, resulting in an annotated dataset including: The second text dataset is matched using each entity data in the entity dataset to obtain the labeled dataset, wherein the labeled dataset includes the entity data matched in the second text dataset and the position of the matched entity data in the second text dataset.

9. The method according to any one of claims 1 to 8, further comprising: In response to a verification operation command for the labeled dataset, the labeled dataset is verified. The output of the labeled dataset includes: the labeled dataset that has been successfully verified.

10. A method for determining a model, comprising: Obtaining a labeled dataset, wherein the labeled dataset is obtained by labeling the text dataset to be labeled in the text dataset using an entity dataset, and includes the entity dataset, wherein the quality assessment result of the entity data in the entity dataset is greater than a quality assessment threshold, and the entity dataset is determined based on candidate entity data in the first text dataset whose frequency of occurrence is higher than a first frequency threshold, and determining the entity dataset based on the candidate entity data in the first text dataset whose frequency of occurrence is higher than the first frequency threshold includes: obtaining target candidate entity data including the candidate entity data from the first text dataset; and determining the entity dataset based on the target candidate entity data in the first text dataset whose frequency of occurrence is higher than a second frequency threshold, wherein the second frequency threshold is less than the first frequency threshold; Supervised training is performed based on the labeled dataset to obtain an entity recognition model, which is used to identify entity data in the text to be recognized.

11. A text data processing apparatus, comprising: The first acquisition unit is used to acquire the first text dataset to be converted from the text dataset; The transformation unit is used to determine the entity dataset based on candidate entity data in the first text dataset whose frequency of occurrence is higher than a first frequency threshold; A labeling unit is configured to, in response to a quality assessment result of entity data in the entity dataset being greater than a quality assessment threshold, label a second text dataset to be labeled in the text dataset using the entity dataset, thereby obtaining a labeled dataset, wherein the labeled dataset includes the entity dataset; The output unit is used to output the labeled dataset, wherein the labeled dataset is used to obtain an entity recognition model through supervised training, and the entity recognition model is used to identify entity data in the text to be recognized; The conversion unit is configured to determine the entity dataset based on candidate entity data in the first text dataset that appears more frequently than a first frequency threshold by performing the following steps: obtaining target candidate entity data including the candidate entity data from the first text dataset; determining the entity dataset based on the target candidate entity data in the first text dataset that appears more frequently than a second frequency threshold, wherein the second frequency threshold is less than the first frequency threshold.

12. A model determining device, comprising: The second acquisition unit is used to acquire a labeled dataset, wherein the labeled dataset is obtained by labeling the text dataset to be labeled in the text dataset using an entity dataset, and includes the entity dataset, wherein the quality assessment result of the entity data in the entity dataset is greater than a quality assessment threshold, and the entity dataset is determined based on candidate entity data in the first text dataset whose frequency of occurrence is higher than a first frequency threshold. Determining the entity dataset based on the candidate entity data in the first text dataset whose frequency of occurrence is higher than the first frequency threshold includes: acquiring target candidate entity data including the candidate entity data from the first text dataset; and determining the entity dataset based on the target candidate entity data in the first text dataset whose frequency of occurrence is higher than a second frequency threshold, wherein the second frequency threshold is less than the first frequency threshold. The training unit is used to perform supervised training based on the labeled dataset to obtain an entity recognition model, wherein the entity recognition model is used to identify entity data in the text to be recognized.

13. An electronic device, comprising: At least one processor; as well as A memory communicatively connected to the at least one processor; wherein, The memory stores instructions executable by the at least one processor, which, when executed by the at least one processor, enables the at least one processor to perform the method of any one of claims 1-10.

14. A non-transitory computer-readable storage medium storing computer instructions, wherein, The computer instructions are used to cause the computer to perform the method according to any one of claims 1-10.

15. A computer program product comprising a computer program that, when executed by a processor, implements the method according to any one of claims 1-10.

Images