A real estate industry quantitative text extraction method, system, device and storage medium

Abstract

The embodiment of the application discloses a real estate industry quantitative text extraction method, system, device and storage medium, first obtain the dialogue text input by the real estate industry real estate consultant; the dialogue text is matched to the label, the dialogue text is marked according to the label matching result, the fuzzy quantifier in the first label text pair is cleaned based on the label in the first label text pair after marking, and the second label text pair is obtained;According to the label type in the second label text pair, the quantitative text in it is standardized, and the standard label text pair is obtained;The standard label text pair is input into the quantitative text extraction labeling model trained in advance, and the quantitative label data pair is obtained;The quantitative label data pair is checked, and the standard quantitative data pair is output according to the checking result.The embodiment of the application realizes the standardization extraction of informal text quantitative information, expands the range of quantitative extraction object, and effectively improves the quantitative extraction efficiency.

Description

Technical Field

[0001] This invention relates to the field of data analysis, specifically to a method, system, device, and storage medium for quantitative text extraction in the real estate industry. Background Technology

[0002] Currently, user profile tags used in the real estate industry mainly rely on business data. This requires extracting quantitative information about users from a large amount of dialogue texts with notes from real estate consultants. However, dialogue texts are informal texts with diverse content and formats, making it difficult to effectively extract the required quantitative information.

[0003] Most existing quantitative text extraction techniques are based on deep learning algorithms, requiring extensive text annotation of domain-specific texts, resulting in high labor costs and low extraction efficiency. Furthermore, current technologies can only extract standardized quantitative information from texts containing standard quantitative information, failing to achieve standardized extraction of quantitative information from informal texts. This limits the scope of current quantitative information extraction techniques. Summary of the Invention

[0004] Therefore, embodiments of the present invention provide a method, system, device and storage medium for quantitative text extraction in the real estate industry, in order to solve the problems of low quantitative extraction efficiency and limited quantitative extraction objects in existing quantitative text extraction technologies.

[0005] To achieve the above objectives, the embodiments of the present invention provide the following technical solutions:

[0006] According to a first aspect of the present invention, a method for quantitative text extraction in the real estate industry is provided, the method comprising:

[0007] Obtain the dialogue text entered by real estate consultants after each visitor meeting;

[0008] The dialogue text is subjected to tag matching, and the dialogue text is labeled according to the tag matching results. The labeled dialogue text is used as the first tag text pair.

[0009] Based on the tags in the first tag text pair, the fuzzy quantifiers in the first tag text pair are cleaned to obtain the second tag text pair;

[0010] Based on the tag type of the tags in the second tag text pair, the quantitative text in the second tag text pair is standardized to obtain a standard tag text pair;

[0011] The standard label text pairs are input into a pre-trained quantitative text extraction and annotation model to obtain quantitative label data pairs;

[0012] The quantitative label data pairs are verified, and the quantitative label data pairs are supplemented according to the verification results to obtain standard quantitative data pairs that have passed the verification.

[0013] Further, the dialogue text is subjected to tag matching, and the dialogue text is labeled according to the tag matching results. The labeled dialogue text is used as the first tag text, including:

[0014] Based on the punctuation marks in the dialogue text, the dialogue text is segmented into short sentences to obtain the segmented short sentence text.

[0015] For each of the short sentence texts, based on a preset word tag library, word tag matching is performed on the short sentence text to determine whether there is a word tag in the preset word tag library that matches the short sentence text;

[0016] If no word tag in the preset word tag library matches the short sentence text, the short sentence text will be discarded.

[0017] If there is a word tag in the preset word tag library that matches the short sentence text, then the short sentence text is annotated using the matched word tag, and the annotated short sentence text is used as the first annotated short sentence text.

[0018] Using the preset word filter tag library corresponding to the word label, the first labeled short sentence text is matched with word filter tags to determine whether there are word filter tags in the preset word filter tag library that match the first labeled short sentence text.

[0019] If no word filter tag in the preset word filter tag library matches the first annotated short sentence text, then the first annotated short sentence text is discarded;

[0020] If there is a word filter tag in the preset word filter tag library that matches the first annotated short sentence text, then the first annotated short sentence text is annotated a second time using the matched word filter tag to obtain the second annotated short sentence text;

[0021] The dialogue text corresponding to the second annotated short sentence text and the two adjacent dialogue texts before and after it are spliced ​​together in order to obtain the target long text;

[0022] Using a preset sliding window, a preset number of short sentences are sequentially extracted from the beginning of the target long text to obtain the slid-extracted text;

[0023] For the sliding text, calculate the vector similarity between the sliding text and each preset text in the preset corpus corresponding to the word filter tag, and obtain the vector similarity calculation result;

[0024] Determine whether the vector similarity calculation result is greater than or equal to a preset similarity threshold;

[0025] If the vector similarity calculation result is greater than or equal to the preset similarity threshold, then the first tag text pair is obtained based on the dialogue text corresponding to the second labeled short sentence text and the corresponding word label;

[0026] If the vector similarity is less than a preset similarity threshold, then it is determined whether the sliding window has slid to the end of the target long text;

[0027] If the sliding window does not slide to the end of the target long text, the sliding window slides at the current position to the end of the target long text by the distance of a short sentence text, and repeats until a preset number of short sentences text are extracted to obtain the slid-trimmed text;

[0028] If the sliding window has slid to the end of the target long text, then the first tag text pair is obtained based on the second annotated short sentence text and the corresponding word filter tag.

[0029] Furthermore, based on the tags in the first tag text pair, the fuzzy quantifiers in the first tag text pair are cleaned to obtain the second tag text pair, including:

[0030] The tags in the first tag text pair are matched with a preset first quantitative tag library to obtain the first matching result;

[0031] Determine whether the first matching result is successful;

[0032] If the first matching result fails, no data cleaning processing will be performed on the first tag text pair;

[0033] If the first matching result is successful, then perform text structure detection on the text in the first tag text pair to obtain the detection result;

[0034] Based on the detection results, determine whether there are fuzzy quantifiers in the text of the first tag text pair;

[0035] If the text in the first tag text pair contains fuzzy quantifiers, then according to the preset fuzzy quantifier cleaning scheme, the fuzzy quantifiers are cleaned to obtain the second tag text pair;

[0036] If the text in the first tag text pair does not contain any fuzzy quantifiers, then the first tag text pair is directly used as the second tag text pair.

[0037] Furthermore, based on the tag type of the tags in the second tag text pair, the quantitative text in the second tag text pair is standardized to obtain standard tag text pairs, including:

[0038] The tags in the second tag text pair are matched with a preset second quantitative tag library to obtain a second matching result;

[0039] Determine whether the second matching result was successful;

[0040] If the second matching result fails, then quantitative text standardization will not be performed on the text in the second tag text pair;

[0041] If the second matching result is successful, then determine whether the matched quantitative label is a time-related label;

[0042] If the quantitative label is the time-related label, then the time text is extracted from the second label text pair by regular expression matching;

[0043] Based on a preset time standard template, the time text is converted into standard time text, and the standard time text is used to replace the time text in the second tag text pair;

[0044] If the quantitative label is not a time-related label, then determine whether the quantitative label is a monetary label;

[0045] If the quantitative label is the monetary label, then the first monetary text is extracted from the second label text pair by regular expression matching;

[0046] Determine whether the amount value in the first amount text conforms to a preset amount range;

[0047] If the amount value in the first amount text does not conform to the preset amount range, then the first amount text will not be used as the second amount text;

[0048] If the amount value in the first amount text conforms to the preset amount range, then the unit text in the first amount text is replaced with the standard amount unit symbol to obtain the second amount text;

[0049] Determine whether there are reduplicated words in the second amount text;

[0050] If the second amount text contains reduplicated words, then the reduplicated words in the second amount text are dereduplicated to obtain the third amount text;

[0051] If the second amount text does not contain any reduplicated words, then the second amount text is directly used as the third amount text;

[0052] Determine whether the third amount text contains a numerical amount text;

[0053] If the third amount text contains an amount value text, then the amount value text is converted into an amount number to obtain the fourth amount text;

[0054] If the third amount text does not contain an amount value text, then the third amount text is directly used as the fourth amount text;

[0055] Use the fourth amount text as the standard amount text, and replace the first amount text in the second tag text pair with the standard amount text.

[0056] If the quantitative label is not a monetary label, then determine whether the quantitative label is an area label;

[0057] If the quantitative label is an area label, then the first area text is extracted from the second label text pair by regular expression matching;

[0058] Determine whether the area value in the first area text conforms to a preset area range;

[0059] If the area value in the first area text does not conform to the preset area range, then the first area text will not be used as the second area text.

[0060] If the area value in the first area text matches the preset area range, then the unit text in the first area text is replaced with the standard area unit symbol to obtain the second area text;

[0061] Determine whether there are reduplicated words in the second area text;

[0062] If the second area text contains reduplicated words, then the reduplicated words in the second area text are dereduplicated to obtain the third area text;

[0063] If there are no reduplicated words in the second area text, then the second area text is directly used as the third area text;

[0064] Determine whether the third area text contains area numerical text;

[0065] If the third area text contains area numerical text, then the area numerical text is converted into area numbers to obtain the fourth area text;

[0066] If the third area text does not contain an area value text, then the third area text is directly used as the fourth area text;

[0067] Use the fourth area text as the standard area text, and replace the first area text in the second label text pair with the standard area text.

[0068] If the quantitative label is not an area label, then determine whether the quantitative label is a ratio label;

[0069] If the quantitative label is a ratio label, then the ratio text is extracted from the second label text pair by regular expression matching;

[0070] The proportional text is standardized and a proportional symbol is added to obtain standard proportional text. The standard proportional text is then used to replace the proportional text in the second label text pair.

[0071] The second tag text pair after quantitative text replacement is used as the standard tag text pair;

[0072] If the quantitative label is not the proportion type label, then the text in the second label text pair will not be subjected to quantitative text standardization.

[0073] Furthermore, the training process of the quantitative text extraction and annotation model includes:

[0074] Use the general domain dialogue corpus as the first training dataset;

[0075] Input the first training dataset into the pre-trained semantic model;

[0076] For each general dialogue text in the first training dataset, random characters in the general dialogue text are masked, and the random characters are predicted based on the context of the masked random characters to obtain the prediction results.

[0077] The predicted difference value is calculated based on the difference between the predicted result and the masked random character.

[0078] Determine whether the predicted difference value reaches a preset difference threshold;

[0079] If the predicted difference value does not reach the preset difference threshold, the parameters of the pre-trained semantic model are adjusted according to the predicted difference value and the mask prediction is performed again.

[0080] If the difference value reaches the preset difference threshold, the pre-trained semantic model is obtained after training is completed.

[0081] Label the real estate dialogue data and use the labeled real estate dialogue data as the second training dataset;

[0082] The second training dataset is input into the pre-trained semantic model after training for secondary training to obtain a quantitative text extraction and annotation model.

[0083] Further, the standard labeled text pairs are input into a pre-trained quantitative text extraction and annotation model to obtain quantitative labeled data pairs, including:

[0084] The standard label text pairs are input into the quantitative text extraction and annotation model to extract quantitative data and corresponding quantitative labels from the standard label text pairs;

[0085] The quantitative data and the corresponding quantitative labels are used as quantitative label data pairs.

[0086] Further, the quantitative label data pairs are validated, and the data pairs are supplemented based on the validation results to obtain validated standard quantitative data pairs, including:

[0087] Determine whether Chinese characters exist in the quantitative data of the quantitative label data pair;

[0088] If Chinese characters are present in the quantitative data, the quantitative label data pair will be discarded.

[0089] If no Chinese characters are present in the quantitative data, then determine whether there are unit symbols in the quantitative data.

[0090] If the quantitative data does not contain a unit symbol, the quantitative data is supplemented with units according to a preset quantitative data template, and the supplemented quantitative label data pair is used as a standard quantitative data pair.

[0091] If the quantitative data contains a unit of measurement symbol, then the quantitative data is determined to be within the preset numerical range based on the preset numerical range corresponding to the quantitative label.

[0092] If the quantitative data does not conform to the preset numerical range, the quantitative label data pair will be discarded;

[0093] If the quantitative data meets the preset numerical range, the quantitative label data pair is used as the standard quantitative data pair.

[0094] According to a second aspect of the present invention, a quantitative text extraction system for the real estate industry is provided, the system comprising:

[0095] The data acquisition module is used to acquire the dialogue text entered by real estate consultants after each visitor meeting;

[0096] The text annotation module is used to perform tag matching on the dialogue text, annotate the dialogue text according to the tag matching results, and use the annotated dialogue text as the first tag text pair.

[0097] The data cleaning module is used to clean the fuzzy quantifiers in the first tag text pair based on the tags in the first tag text pair, so as to obtain the second tag text pair.

[0098] The quantitative text standardization module is used to standardize the quantitative text in the second label text pair according to the label type of the label in the second label text pair, so as to obtain a standard label text pair.

[0099] The quantitative text extraction module is used to input the standard label text pairs into a pre-trained quantitative text extraction and annotation model to obtain quantitative label data pairs;

[0100] The data verification module is used to verify the quantitative label data pairs, and to complete the quantitative label data pairs according to the verification results to obtain standard quantitative data pairs that have passed the verification.

[0101] Further, the dialogue text is subjected to tag matching, and the dialogue text is labeled according to the tag matching results. The labeled dialogue text is used as the first tag text, including:

[0102] Based on the punctuation marks in the dialogue text, the dialogue text is segmented into short sentences to obtain the segmented short sentence text.

[0103] For each of the short sentence texts, based on a preset word tag library, word tag matching is performed on the short sentence text to determine whether there is a word tag in the preset word tag library that matches the short sentence text;

[0104] If no word tag in the preset word tag library matches the short sentence text, the short sentence text will be discarded.

[0105] If there is a word tag in the preset word tag library that matches the short sentence text, then the short sentence text is annotated using the matched word tag, and the annotated short sentence text is used as the first annotated short sentence text.

[0106] Using the preset word filter tag library corresponding to the word label, the first labeled short sentence text is matched with word filter tags to determine whether there are word filter tags in the preset word filter tag library that match the first labeled short sentence text.

[0107] If no word filter tag in the preset word filter tag library matches the first annotated short sentence text, then the first annotated short sentence text is discarded;

[0108] If there is a word filter tag in the preset word filter tag library that matches the first annotated short sentence text, then the first annotated short sentence text is annotated a second time using the matched word filter tag to obtain the second annotated short sentence text;

[0109] The dialogue text corresponding to the second annotated short sentence text and the two adjacent dialogue texts before and after it are spliced ​​together in order to obtain the target long text;

[0110] Using a preset sliding window, a preset number of short sentences are sequentially extracted from the beginning of the target long text to obtain the slid-extracted text;

[0111] For the sliding text, calculate the vector similarity between the sliding text and each preset text in the preset corpus corresponding to the word filter tag, and obtain the vector similarity calculation result;

[0112] Determine whether the vector similarity calculation result is greater than or equal to a preset similarity threshold;

[0113] If the vector similarity calculation result is greater than or equal to the preset similarity threshold, then the first tag text pair is obtained based on the dialogue text corresponding to the second labeled short sentence text and the corresponding word label;

[0114] If the vector similarity is less than a preset similarity threshold, then it is determined whether the sliding window has slid to the end of the target long text;

[0115] If the sliding window does not slide to the end of the target long text, the sliding window slides at the current position to the end of the target long text by the distance of a short sentence text, and repeats until a preset number of short sentences text are extracted to obtain the slid-trimmed text;

[0116] If the sliding window has slid to the end of the target long text, then the first tag text pair is obtained based on the second annotated short sentence text and the corresponding word filter tag.

[0117] Furthermore, based on the tags in the first tag text pair, the fuzzy quantifiers in the first tag text pair are cleaned to obtain the second tag text pair, including:

[0118] The tags in the first tag text pair are matched with a preset first quantitative tag library to obtain the first matching result;

[0119] Determine whether the first matching result is successful;

[0120] If the first matching result fails, no data cleaning processing will be performed on the first tag text pair;

[0121] If the first matching result is successful, then perform text structure detection on the text in the first tag text pair to obtain the detection result;

[0122] Based on the detection results, determine whether there are fuzzy quantifiers in the text of the first tag text pair;

[0123] If the text in the first tag text pair contains fuzzy quantifiers, then according to the preset fuzzy quantifier cleaning scheme, the fuzzy quantifiers are cleaned to obtain the second tag text pair;

[0124] If the text in the first tag text pair does not contain any fuzzy quantifiers, then the first tag text pair is directly used as the second tag text pair.

[0125] Furthermore, based on the tag type of the tags in the second tag text pair, the quantitative text in the second tag text pair is standardized to obtain standard tag text pairs, including:

[0126] The tags in the second tag text pair are matched with a preset second quantitative tag library to obtain a second matching result;

[0127] Determine whether the second matching result was successful;

[0128] If the second matching result fails, then quantitative text standardization will not be performed on the text in the second tag text pair;

[0129] If the second matching result is successful, then determine whether the matched quantitative label is a time-related label;

[0130] If the quantitative label is the time-related label, then the time text is extracted from the second label text pair by regular expression matching;

[0131] Based on a preset time standard template, the time text is converted into standard time text, and the standard time text is used to replace the time text in the second tag text pair;

[0132] If the quantitative label is not a time-related label, then determine whether the quantitative label is a monetary label;

[0133] If the quantitative label is the monetary label, then the first monetary text is extracted from the second label text pair by regular expression matching;

[0134] Determine whether the amount value in the first amount text conforms to a preset amount range;

[0135] If the amount value in the first amount text does not conform to the preset amount range, then the first amount text will not be used as the second amount text;

[0136] If the amount value in the first amount text conforms to the preset amount range, then the unit text in the first amount text is replaced with the standard amount unit symbol to obtain the second amount text;

[0137] Determine whether there are reduplicated words in the second amount text;

[0138] If the second amount text contains reduplicated words, then the reduplicated words in the second amount text are dereduplicated to obtain the third amount text;

[0139] If the second amount text does not contain any reduplicated words, then the second amount text is directly used as the third amount text;

[0140] Determine whether the third amount text contains a numerical amount text;

[0141] If the third amount text contains an amount value text, then the amount value text is converted into an amount number to obtain the fourth amount text;

[0142] If the third amount text does not contain an amount value text, then the third amount text is directly used as the fourth amount text;

[0143] Use the fourth amount text as the standard amount text, and replace the first amount text in the second tag text pair with the standard amount text.

[0144] If the quantitative label is not a monetary label, then determine whether the quantitative label is an area label;

[0145] If the quantitative label is an area label, then the first area text is extracted from the second label text pair by regular expression matching;

[0146] Determine whether the area value in the first area text conforms to a preset area range;

[0147] If the area value in the first area text does not conform to the preset area range, then the first area text will not be used as the second area text.

[0148] If the area value in the first area text matches the preset area range, then the unit text in the first area text is replaced with the standard area unit symbol to obtain the second area text;

[0149] Determine whether there are reduplicated words in the second area text;

[0150] If the second area text contains reduplicated words, then the reduplicated words in the second area text are dereduplicated to obtain the third area text;

[0151] If there are no reduplicated words in the second area text, then the second area text is directly used as the third area text;

[0152] Determine whether the third area text contains area numerical text;

[0153] If the third area text contains area numerical text, then the area numerical text is converted into area numbers to obtain the fourth area text;

[0154] If the third area text does not contain an area value text, then the third area text is directly used as the fourth area text;

[0155] Use the fourth area text as the standard area text, and replace the first area text in the second label text pair with the standard area text.

[0156] If the quantitative label is not an area label, then determine whether the quantitative label is a ratio label;

[0157] If the quantitative label is a ratio label, then the ratio text is extracted from the second label text pair by regular expression matching;

[0158] The proportional text is standardized and a proportional symbol is added to obtain standard proportional text. The standard proportional text is then used to replace the proportional text in the second label text pair.

[0159] The second tag text pair after quantitative text replacement is used as the standard tag text pair;

[0160] If the quantitative label is not the proportion type label, then the text in the second label text pair will not be subjected to quantitative text standardization.

[0161] Furthermore, the training process of the quantitative text extraction and annotation model includes:

[0162] Use the general domain dialogue corpus as the first training dataset;

[0163] Input the first training dataset into the pre-trained semantic model;

[0164] For each general dialogue text in the first training dataset, random characters in the general dialogue text are masked, and the random characters are predicted based on the context of the masked random characters to obtain the prediction results.

[0165] The predicted difference value is calculated based on the difference between the predicted result and the masked random character.

[0166] Determine whether the predicted difference value reaches a preset difference threshold;

[0167] If the predicted difference value does not reach the preset difference threshold, the parameters of the pre-trained semantic model are adjusted according to the predicted difference value and the mask prediction is performed again.

[0168] If the difference value reaches the preset difference threshold, the pre-trained semantic model is obtained after training is completed.

[0169] Label the real estate dialogue data and use the labeled real estate dialogue data as the second training dataset;

[0170] The second training dataset is input into the pre-trained semantic model after training for secondary training to obtain a quantitative text extraction and annotation model.

[0171] Further, the standard labeled text pairs are input into a pre-trained quantitative text extraction and annotation model to obtain quantitative labeled data pairs, including:

[0172] The standard label text pairs are input into the quantitative text extraction and annotation model to extract quantitative data and corresponding quantitative labels from the standard label text pairs;

[0173] The quantitative data and the corresponding quantitative labels are used as quantitative label data pairs.

[0174] Further, the quantitative label data pairs are validated, and the data pairs are supplemented based on the validation results to obtain validated standard quantitative data pairs, including:

[0175] Determine whether Chinese characters exist in the quantitative data of the quantitative label data pair;

[0176] If Chinese characters are present in the quantitative data, the quantitative label data pair will be discarded.

[0177] If no Chinese characters are present in the quantitative data, then determine whether there are unit symbols in the quantitative data.

[0178] If the quantitative data does not contain a unit symbol, the quantitative data is supplemented with units according to a preset quantitative data template, and the supplemented quantitative label data pair is used as a standard quantitative data pair.

[0179] If the quantitative data contains a unit of measurement symbol, then the quantitative data is determined to be within the preset numerical range based on the preset numerical range corresponding to the quantitative label.

[0180] If the quantitative data does not conform to the preset numerical range, the quantitative label data pair will be discarded;

[0181] If the quantitative data meets the preset numerical range, the quantitative label data pair is used as the standard quantitative data pair.

[0182] According to a third aspect of the present invention, a quantitative text extraction device for the real estate industry is provided, the device comprising: a processor and a memory;

[0183] The memory is used to store one or more program instructions;

[0184] The processor is configured to run one or more program instructions to perform the steps of a quantitative text extraction method for the real estate industry as described in any of the preceding claims.

[0185] According to a fourth aspect of the present invention, a computer-readable storage medium is provided, on which a computer program is stored, wherein when executed by a processor, the computer program implements the steps of a quantitative text extraction method for the real estate industry as described in any of the preceding claims.

[0186] The embodiments of the present invention have the following advantages:

[0187] This invention discloses a method, system, device, and storage medium for quantitative text extraction in the real estate industry. First, it acquires dialogue text input by a real estate consultant. The dialogue text is then tagged with labels, and annotated based on the matching results. Based on the labels in the first tagged text pair, fuzzy quantifiers in the first tagged text pair are cleaned to obtain a second tagged text pair. According to the label type in the second tagged text pair, the quantitative text is standardized to obtain a standard tagged text pair. The standard tagged text pair is input into a pre-trained quantitative text extraction and annotation model to obtain quantitative tagged data pairs. The quantitative tagged data pairs are then validated, and standard quantitative data pairs are output based on the validation results. This invention achieves standardized extraction of quantitative information from informal text, expands the scope of quantitative extraction objects, and effectively improves quantitative extraction efficiency. Attached Figure Description

[0188] To more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely exemplary, and those skilled in the art can derive other embodiments based on the provided drawings without creative effort.

[0189] The structures, proportions, sizes, etc. illustrated in this specification are only for the purpose of assisting those skilled in the art in understanding and reading the content disclosed herein, and are not intended to limit the conditions under which the present invention can be implemented. Therefore, they have no substantial technical significance. Any modifications to the structure, changes in the proportions, or adjustments to the size, without affecting the effects and objectives that the present invention can produce, should still fall within the scope of the technical content disclosed in the present invention.

[0190] Figure 1 A schematic diagram of the logical structure of a quantitative text extraction system for the real estate industry provided in an embodiment of the present invention;

[0191] Figure 2 A flowchart illustrating a quantitative text extraction method for the real estate industry provided in this embodiment of the invention;

[0192] Figure 3 A schematic diagram illustrating the process of dialog text tagging provided in an embodiment of the present invention;

[0193] Figure 4 A schematic diagram of the label text data cleaning process provided in an embodiment of the present invention;

[0194] Figure 5 A flowchart illustrating the quantitative text standardization process provided in this embodiment of the invention;

[0195] Figure 6 This is a schematic diagram illustrating the training and application process of the quantitative text extraction and annotation model provided in an embodiment of the present invention.

[0196] Figure 7 This is a schematic diagram of the quantitative label data verification process provided in an embodiment of the present invention. Detailed Implementation

[0197] The following specific embodiments illustrate the implementation of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0198] refer to Figure 1 This invention provides a quantitative text extraction system for the real estate industry, which specifically includes: a data acquisition module 1, a text annotation module 2, a data cleaning module 3, a quantitative text standardization module 4, a quantitative text extraction module 5, and a data verification module 6.

[0199] Furthermore, the data acquisition module 1 is used to acquire the dialogue text entered by real estate consultants after each visitor reception; the text annotation module 2 is used to perform tag matching on the dialogue text, and annotate the dialogue text according to the tag matching results, using the annotated dialogue text as the first tag text pair; the data cleaning module 3 is used to clean the fuzzy quantifiers in the first tag text pair based on the tags in the first tag text pair, to obtain the second tag text pair; the quantitative text standardization module 4 is used to standardize the quantitative text in the second tag text pair according to the tag type of the tags in the second tag text pair, to obtain the standard tag text pair; the quantitative text extraction module 5 is used to input the standard tag text pair into a pre-trained quantitative text extraction and annotation model, to obtain the quantitative tag data pair; and the data verification module 6 is used to verify the quantitative tag data pair, and complete the quantitative tag data pair according to the verification results, to obtain the verified standard quantitative data pair.

[0200] This invention discloses a quantitative text extraction system for the real estate industry. First, it acquires dialogue text input by real estate consultants. The dialogue text is then tagged with labels, and annotated based on the matching results. Based on the labels in the first tagged text pair, fuzzy quantifiers in the first tagged text pair are cleaned to obtain a second tagged text pair. According to the label type in the second tagged text pair, the quantitative text is standardized to obtain a standard tagged text pair. The standard tagged text pair is then input into a pre-trained quantitative text extraction and annotation model to obtain quantitative tagged data pairs. Finally, the quantitative tagged data pairs are validated, and standard quantitative data pairs are output based on the validation results. This invention achieves standardized extraction of quantitative information from informal text, expands the scope of quantitative extraction objects, and effectively improves quantitative extraction efficiency.

[0201] Corresponding to the aforementioned quantitative text extraction system for the real estate industry, this invention also discloses a method for quantitative text extraction in the real estate industry. The following details a method for quantitative text extraction in the real estate industry disclosed in this invention, in conjunction with the aforementioned quantitative text extraction system.

[0202] refer to Figure 2 The following describes the specific steps of a quantitative text extraction method for the real estate industry provided by an embodiment of the present invention.

[0203] The data acquisition module 1 acquires the dialogue text entered by real estate consultants after each visitor meeting.

[0204] The above steps specifically include: obtaining the notes left by real estate consultants in the visitor system after each visit, including the dialogue text between the consultant and the client.

[0205] The text annotation module 2 performs tag matching on the dialogue text, annotates the dialogue text based on the tag matching results, and uses the annotated dialogue text as the first tag text pair.

[0206] refer to Figure 3 The above steps specifically include: First, preprocessing the dialogue text by segmenting it into short sentences based on punctuation marks; for each short sentence, performing word tag matching based on a pre-defined word tag library to determine if a matching word tag exists in the library; if no matching word tag exists, the short sentence is discarded; if a matching word tag exists, the short sentence is annotated using the word tag, and the annotated short sentence is then processed. The first annotated short sentence text is generated by: using a preset word filter tag library corresponding to the annotated word labels, performing word filter tag matching on the first annotated short sentence text; determining whether there is a word filter tag in the preset word filter tag library that matches the first annotated short sentence text; if there is no word filter tag in the preset word filter tag library that matches the first annotated short sentence text, then the first annotated short sentence text is discarded; if there is a word filter tag in the preset word filter tag library that matches the first annotated short sentence text, then the first annotated short sentence text is annotated a second time using the word filter tags to obtain the second annotated short sentence text; and then combining the dialogue text corresponding to the second annotated short sentence text with the preceding and following dialogue texts. Two adjacent dialogue texts are concatenated sequentially to obtain the target long text. A preset sliding window is used to sequentially extract a preset number of short sentences from the beginning of the target long text, resulting in truncated text. For each truncated text, the vector similarity between the truncated text and each preset text in the preset corpus corresponding to the aforementioned word filter tags is calculated, yielding the vector similarity calculation result. It is then determined whether the vector similarity calculation result is greater than or equal to a preset similarity threshold. If the vector similarity calculation result is greater than or equal to the preset similarity threshold, the dialogue text corresponding to the second labeled short sentence text is used as the key. The key of the value pair is used as the corresponding word label as the value of the key-value pair, resulting in the first label text pair in key-value pair form. If the vector similarity is less than the preset similarity threshold, it is determined whether the sliding window has slid to the end of the target long text. If the sliding window has not slid to the end of the target long text, the sliding window slides to the end of the target long text at the current position by the distance of a short sentence text, and repeats until a preset number of short sentence texts are extracted to obtain the truncated text. If the sliding window has slid to the end of the target long text, the first label text pair in key-value pair form is obtained based on the second labeled short sentence text and the corresponding word filter label.

[0207] This invention provides a qualitative labeling method for dialogue texts in the real estate industry. Based on the labeling results, text filtering is performed to remove useless dialogue texts and extract highly relevant texts from the dialogue texts, effectively improving the efficiency of subsequent quantitative text standardization and quantitative text extraction.

[0208] The data cleaning module 3 cleans the fuzzy quantifiers in the first tag text pair based on the tags in the first tag text pair to obtain the second tag text pair.

[0209] refer to Figure 4 The above steps specifically include: first, matching the tags in the first tag text pair with a preset first quantitative tag library to obtain a first matching result; determining whether the first matching result is successful; if the first matching result fails, no data cleaning processing is performed on the first tag text pair; if the first matching result is successful, text structure detection is performed on the text in the first tag text pair to obtain a detection result; based on the detection result, determining whether there are fuzzy quantifiers in the text of the first tag text pair, where the fuzzy quantifiers are diversified and vague quantitative textual expressions, such as "more than one million" or "seventy percent"; if there are fuzzy quantifiers in the text of the first tag text pair, regular expression data cleaning processing is performed on the fuzzy quantifiers according to a preset fuzzy quantifier cleaning scheme, for example, cleaning fuzzy quantifiers such as "more than 100,000", "ninety-odd square meters", and "about two million and fifty thousand" into "100,000", "ninety square meters", and "two million and fifty thousand", to obtain the second tag text pair after cleaning; if there are no fuzzy quantifiers in the text of the first tag text pair, the first tag text pair is directly used as the second tag text pair.

[0210] This invention detects fuzzy quantifiers in the first tag text through text structure detection, specifies a fuzzy quantifier cleaning scheme according to business requirements, and then uses the cleaning scheme to clean the fuzzy quantifiers, transforming the fuzzy quantitative text into definite quantitative text. The cleaned quantitative text facilitates the improvement of the efficiency and accuracy of subsequent standardization processing.

[0211] The quantitative text standardization module 4 standardizes the quantitative text in the second label text pair according to the label type of the label in the second label text pair, and obtains the standard label text pair.

[0212] refer to Figure 5The above steps specifically include: Quantitative texts in the real estate industry mainly include four categories: time, amount, area, and proportion. First, the tags in the second tag text pair are matched with a preset second quantitative tag library to obtain a second matching result; it is then determined whether the second matching result is successful; if the second matching result fails, it means that the second tag text pair does not need quantitative standardization, and no quantitative text standardization processing is performed on the text in the second tag text pair; if the second matching result is successful, then quantitative standardization processing is performed on the second tag text pair according to the tag type of the matched quantitative tag.

[0213] If the quantitative label is a time-related label, the time text is extracted from the second label text pair using regular expression matching; based on the preset time standard template, the time text is converted into standard time text, and the standard time text is used to replace the time text in the second label text pair.

[0214] If the quantitative label is an amount label, then the first amount text is extracted from the second label text pair using regular expression matching; it is determined whether the amount value in the first amount text conforms to a preset amount range; if the amount value in the first amount text does not conform to the preset amount range, then the first amount text is not used as the second amount text; if the amount value in the first amount text conforms to the preset amount range, then the unit text in the first amount text is replaced with a standard amount unit symbol to obtain the second amount text; it is determined whether there are reduplicated words in the second amount text; if there are reduplicated words in the second amount text, then the reduplicated words in the second amount text are dereduplicated to obtain the third amount text; if there are no reduplicated words in the second amount text, then the second amount text is directly used as the third amount text; it is determined whether there are amount value texts in the third amount text; if there are amount value texts in the third amount text, then the amount value texts are converted into amount numbers to obtain the fourth amount text; if there are no amount value texts in the third amount text, then the third amount text is directly used as the fourth amount text; the fourth amount text is used as the standard amount text, and the first amount text in the second label text pair is replaced with the standard amount text.

[0215] If the quantitative label is an area-type label, then the first area text is extracted from the second label text pair using regular expression matching; it is determined whether the area value in the first area text conforms to a preset area range; if the area value in the first area text does not conform to the preset area range, then the first area text is not used as the second area text; if the area value in the first area text conforms to the preset area range, then the unit text in the first area text is replaced with the standard area unit symbol to obtain the second area text; it is determined whether there are reduplicated words in the second area text; if there are reduplicated words in the second area text, then the reduplicated words in the second area text are deredundant to obtain the third area text; if there are no reduplicated words in the second area text, then the second area text is directly used as the third area text; it is determined whether there is area value text in the third area text; if there is area value text in the third area text, then the area value text is converted into area numbers to obtain the fourth area text; if there is no area value text in the third area text, then the third area text is directly used as the fourth area text; the fourth area text is used as the standard area text, and the first area text in the second label text pair is replaced with the standard area text.

[0216] If the quantitative label is a ratio label, the ratio text is extracted from the second label text pair using regular expression matching; the ratio text is standardized and a ratio symbol is appended to obtain the standard ratio text, which is then used to replace the ratio text in the second label text pair.

[0217] The second tag text pair after quantitative text replacement is used as the standard tag text pair.

[0218] According to the quantitative text type in the second tag text pair, the present invention adopts a template-based standardization method for time-type quantitative text and a rule-based standardization method for amount-type, area-type, and ratio-type quantitative text, thereby converting fuzzy and diversified quantitative text into standardized quantitative text. The standardized quantitative text can improve the accuracy of subsequent quantitative information extraction.

[0219] The quantitative text extraction module 5 inputs the standard label text pairs into the pre-trained quantitative text extraction and annotation model to obtain quantitative label data pairs.

[0220] refer to Figure 6The above steps specifically include a model training process and a model application process. The model training process specifically includes: using a large amount of publicly available general-domain dialogue corpus as the first training dataset; inputting the first training dataset into a pre-trained semantic model; the pre-trained semantic model uses the Roberta model as a framework and the MLM task as the target for machine learning. First, for each general-domain dialogue text in the first training dataset, random characters in the general-domain dialogue text are masked. Based on the context of the masked random characters, the random characters are predicted to obtain the prediction results. Based on the difference between the prediction results and the masked random characters, the prediction difference value is calculated. It is then determined whether the prediction difference value reaches a preset difference threshold. If the predicted difference value does not reach the preset difference threshold, the parameters of the pre-trained semantic model are adjusted according to the predicted difference value, and the masking process is repeated to perform mask prediction again. If the difference value reaches the preset difference threshold, the pre-trained semantic model is obtained. Then, a small amount of manpower is used to label the dialogue data in the real estate field. The labeled real estate dialogue data is used as the second training dataset and input into the pre-trained semantic model for secondary training. The output layer of the model is reconstructed using softmax to obtain a quantitative text extraction and labeling model. This quantitative text extraction and labeling model can extract quantitative data and the corresponding quantitative labels from the text and output the quantitative data and quantitative labels as key-value pairs.

[0221] The model application process specifically includes: inputting standard label text pairs into the quantitative text extraction and annotation model, extracting quantitative data and corresponding quantitative labels from the standard label text pairs, and using the quantitative data and corresponding quantitative labels as quantitative label data pairs.

[0222] In this embodiment of the invention, the model is trained twice using a general domain dialogue corpus and a real estate domain dialogue corpus, respectively, to obtain a quantitative text extraction and annotation model. The model can extract quantitative information and corresponding quantitative labels.

[0223] The data verification module 6 verifies the quantitative label data pairs and completes the quantitative label data pairs according to the verification results to obtain the standard quantitative data pairs that have passed the verification.

[0224] refer to Figure 7The above steps specifically include: First, determining whether Chinese characters exist in the quantitative data of the quantitative label data pair; if Chinese characters exist in the quantitative data, the quantitative label data pair is discarded directly; if no Chinese characters exist in the quantitative data, determining whether unit symbols exist in the quantitative data; if no unit symbols exist in the quantitative data, completing the units in the quantitative data according to the preset quantitative data template, and using the completed quantitative label data pair as the standard quantitative data pair; if unit symbols exist in the quantitative data, determining whether the quantitative data conforms to the preset numerical range corresponding to the quantitative label; if the quantitative data does not conform to the preset numerical range, the quantitative label data pair is also discarded; if the quantitative data conforms to the preset numerical range, the quantitative label data pair is used as the standard quantitative data pair.

[0225] This invention enhances the reliability of the output standard quantitative data pairs by verifying the extracted standard quantitative data pairs, supplementing incomplete quantitative data based on the verification results, and discarding unqualified standard quantitative data pairs.

[0226] This invention discloses a method for quantitative text extraction in the real estate industry. First, it acquires dialogue text input by a real estate consultant. The dialogue text is then tagged with labels, and annotated based on the matching results. Based on the labels in the first tagged text pair, fuzzy quantifiers in the first tagged text pair are cleaned to obtain a second tagged text pair. According to the label type in the second tagged text pair, the quantitative text is standardized to obtain a standard tagged text pair. The standard tagged text pair is then input into a pre-trained quantitative text extraction and annotation model to obtain quantitative tagged data pairs. Finally, the quantitative tagged data pairs are validated, and standard quantitative data pairs are output based on the validation results. This invention achieves standardized extraction of quantitative information from informal text, expands the scope of quantitative extraction objects, and effectively improves the efficiency of quantitative extraction.

[0227] In addition, embodiments of the present invention also provide a quantitative text extraction device for the real estate industry, the device comprising: a processor and a memory; the memory for storing one or more program instructions; the processor for running one or more program instructions to perform the steps of a quantitative text extraction method for the real estate industry as described in any of the preceding embodiments.

[0228] In addition, embodiments of the present invention also provide a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the quantitative text extraction method for the real estate industry as described in any of the preceding embodiments.

[0229] In this embodiment of the invention, the processor can be an integrated circuit chip with signal processing capabilities. The processor can be a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0230] The various methods, steps, and logic diagrams disclosed in the embodiments of this invention can be implemented or executed. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the methods disclosed in the embodiments of this invention can be directly implemented by a hardware decoding processor, or implemented by a combination of hardware and software modules in the decoding processor. The software modules can reside in random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, or other mature storage media in the art. The processor reads information from the storage medium and, in conjunction with its hardware, completes the steps of the above methods.

[0231] The storage medium can be memory, such as volatile memory or non-volatile memory, or may include both volatile and non-volatile memory.

[0232] Among them, non-volatile memory can be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), or flash memory.

[0233] Volatile memory can be random access memory (RAM), which is used as an external cache. By way of example, but not limitation, many forms of RAM are available, such as static random access memory (SRAM), dynamic random access memory (DRAM), synchronous dynamic random access memory (SDRAM), double data rate synchronous dynamic random access memory (DDRSDRAM), enhanced synchronous dynamic random access memory (ESDRAM), synchronous linked dynamic random access memory (Synchlink DRAM, SLDRAM), and direct memory bus RAM (DRRAM).

[0234] The storage media described in the embodiments of the present invention are intended to include, but are not limited to, these and any other suitable types of memory.

[0235] Those skilled in the art will recognize that, in one or more of the examples above, the functions described in this invention can be implemented using a combination of hardware and software. When applied as software, the corresponding functions can be stored in a computer-readable medium or transmitted as one or more instructions or code on a computer-readable medium. Computer-readable media include computer storage media and communication media, wherein communication media include any medium that facilitates the transmission of computer programs from one place to another. Storage media can be any available medium that can be accessed by a general-purpose or special-purpose computer.

[0236] Although the present invention has been described in detail above with general descriptions and specific embodiments, modifications or improvements can be made to it, which will be obvious to those skilled in the art. Therefore, all such modifications or improvements made without departing from the spirit of the present invention fall within the scope of protection claimed by the present invention.

Claims

1. A method for quantitative text extraction in the real estate industry, characterized in that, The method includes: Obtain the dialogue text entered by real estate consultants after each visitor meeting; The dialogue text is subjected to tag matching, and the dialogue text is labeled according to the tag matching results. The labeled dialogue text is used as the first tag text pair. Based on the tags in the first tag text pair, the fuzzy quantifiers in the first tag text pair are cleaned to obtain the second tag text pair; Based on the tag type of the tags in the second tag text pair, the quantitative text in the second tag text pair is standardized to obtain a standard tag text pair; The standard label text pairs are input into a pre-trained quantitative text extraction and annotation model to obtain quantitative label data pairs; The quantitative label data pairs are verified, and the quantitative label data pairs are completed according to the verification results to obtain standard quantitative data pairs that have passed the verification. The dialogue text is subjected to tag matching, and the dialogue text is labeled according to the tag matching results. The labeled dialogue text is used as the first tag text, including: Based on the punctuation marks in the dialogue text, the dialogue text is segmented into short sentences to obtain the segmented short sentence text. For each of the short sentence texts, based on a preset word tag library, word tag matching is performed on the short sentence text to determine whether there is a word tag in the preset word tag library that matches the short sentence text; If no word tag in the preset word tag library matches the short sentence text, the short sentence text will be discarded. If there is a word tag in the preset word tag library that matches the short sentence text, then the short sentence text is annotated using the matched word tag, and the annotated short sentence text is used as the first annotated short sentence text. Using the preset word filter tag library corresponding to the word label, the first labeled short sentence text is matched with word filter tags to determine whether there are word filter tags in the preset word filter tag library that match the first labeled short sentence text. If no word filter tag in the preset word filter tag library matches the first annotated short sentence text, then the first annotated short sentence text is discarded; If there is a word filter tag in the preset word filter tag library that matches the first annotated short sentence text, then the first annotated short sentence text is annotated a second time using the matched word filter tag to obtain the second annotated short sentence text; The dialogue text corresponding to the second annotated short sentence text and the two adjacent dialogue texts before and after it are spliced ​​together in order to obtain the target long text; Using a preset sliding window, a preset number of short sentences are sequentially extracted from the beginning of the target long text to obtain the slid-extracted text; For the sliding text, calculate the vector similarity between the sliding text and each preset text in the preset corpus corresponding to the word filter tag, and obtain the vector similarity calculation result; Determine whether the vector similarity calculation result is greater than or equal to a preset similarity threshold; If the vector similarity calculation result is greater than or equal to the preset similarity threshold, then the first tag text pair is obtained based on the dialogue text corresponding to the second labeled short sentence text and the corresponding word label; If the vector similarity is less than a preset similarity threshold, then it is determined whether the sliding window has slid to the end of the target long text; If the sliding window does not slide to the end of the target long text, the sliding window slides at the current position to the end of the target long text by the distance of a short sentence text, and repeats until a preset number of short sentences text are extracted to obtain the slid-trimmed text; If the sliding window has slid to the end of the target long text, then the first tag text pair is obtained based on the second annotated short sentence text and the corresponding word filter tag; Based on the tags in the first tag text pair, the fuzzy quantifiers in the first tag text pair are cleaned to obtain the second tag text pair, including: The tags in the first tag text pair are matched with a preset first quantitative tag library to obtain the first matching result; Determine whether the first matching result is successful; If the first matching result fails, no data cleaning processing will be performed on the first tag text pair; If the first matching result is successful, then perform text structure detection on the text in the first tag text pair to obtain the detection result; Based on the detection results, determine whether there are fuzzy quantifiers in the text of the first tag text pair; If the text in the first tag text pair contains fuzzy quantifiers, then according to the preset fuzzy quantifier cleaning scheme, the fuzzy quantifiers are cleaned to obtain the second tag text pair; If the text in the first tag text pair does not contain any fuzzy quantifiers, then the first tag text pair is directly used as the second tag text pair.

2. The method for quantitative text extraction in the real estate industry as described in claim 1, characterized in that, Based on the tag type of the tags in the second tag text pair, the quantitative text in the second tag text pair is standardized to obtain standard tag text pairs, including: The tags in the second tag text pair are matched with a preset second quantitative tag library to obtain a second matching result; Determine whether the second matching result was successful; If the second matching result fails, then quantitative text standardization will not be performed on the text in the second tag text pair; If the second matching result is successful, then determine whether the matched quantitative label is a time-related label; If the quantitative label is the time-related label, then the time text is extracted from the second label text pair by regular expression matching; Based on a preset time standard template, the time text is converted into standard time text, and the standard time text is used to replace the time text in the second tag text pair; If the quantitative label is not a time-related label, then determine whether the quantitative label is a monetary label; If the quantitative label is the monetary label, then the first monetary text is extracted from the second label text pair by regular expression matching; Determine whether the amount value in the first amount text conforms to a preset amount range; If the amount value in the first amount text does not conform to the preset amount range, then the first amount text will not be used as the second amount text; If the amount value in the first amount text conforms to the preset amount range, then the unit text in the first amount text is replaced with the standard amount unit symbol to obtain the second amount text; Determine whether there are reduplicated words in the second amount text; If the second amount text contains reduplicated words, then the reduplicated words in the second amount text are dereduplicated to obtain the third amount text; If the second amount text does not contain any reduplicated words, then the second amount text is directly used as the third amount text; Determine whether the third amount text contains a numerical amount text; If the third amount text contains an amount value text, then the amount value text is converted into an amount number to obtain the fourth amount text; If the third amount text does not contain an amount value text, then the third amount text is directly used as the fourth amount text; Use the fourth amount text as the standard amount text, and replace the first amount text in the second tag text pair with the standard amount text. If the quantitative label is not a monetary label, then determine whether the quantitative label is an area label; If the quantitative label is an area label, then the first area text is extracted from the second label text pair by regular expression matching; Determine whether the area value in the first area text conforms to a preset area range; If the area value in the first area text does not conform to the preset area range, then the first area text will not be used as the second area text. If the area value in the first area text matches the preset area range, then the unit text in the first area text is replaced with the standard area unit symbol to obtain the second area text; Determine whether there are reduplicated words in the second area text; If the second area text contains reduplicated words, then the reduplicated words in the second area text are dereduplicated to obtain the third area text; If there are no reduplicated words in the second area text, then the second area text is directly used as the third area text; Determine whether the third area text contains area numerical text; If the third area text contains area numerical text, then the area numerical text is converted into area numbers to obtain the fourth area text; If the third area text does not contain an area value text, then the third area text is directly used as the fourth area text; Use the fourth area text as the standard area text, and replace the first area text in the second label text pair with the standard area text. If the quantitative label is not an area label, then determine whether the quantitative label is a ratio label; If the quantitative label is a ratio label, then the ratio text is extracted from the second label text pair by regular expression matching; The proportional text is standardized and a proportional symbol is added to obtain standard proportional text. The standard proportional text is then used to replace the proportional text in the second label text pair. The second tag text pair after quantitative text replacement is used as the standard tag text pair; If the quantitative label is not the proportion type label, then the text in the second label text pair will not be subjected to quantitative text standardization.

3. The method for quantitative text extraction in the real estate industry as described in claim 2, characterized in that, The training process of the quantitative text extraction and annotation model includes: Use the general domain dialogue corpus as the first training dataset; Input the first training dataset into the pre-trained semantic model; For each general dialogue text in the first training dataset, random characters in the general dialogue text are masked, and the random characters are predicted based on the context of the masked random characters to obtain the prediction results. The predicted difference value is calculated based on the difference between the predicted result and the masked random character. Determine whether the predicted difference value reaches a preset difference threshold; If the predicted difference value does not reach the preset difference threshold, the parameters of the pre-trained semantic model are adjusted according to the predicted difference value and the mask prediction is performed again. If the difference value reaches the preset difference threshold, the pre-trained semantic model is obtained after training is completed. Label the real estate dialogue data and use the labeled real estate dialogue data as the second training dataset; The second training dataset is input into the pre-trained semantic model after training for secondary training to obtain a quantitative text extraction and annotation model.

4. The method for quantitative text extraction in the real estate industry as described in claim 3, characterized in that, The standard label text pairs are input into a pre-trained quantitative text extraction and annotation model to obtain quantitative label data pairs, including: The standard label text pairs are input into the quantitative text extraction and annotation model to extract quantitative data and corresponding quantitative labels from the standard label text pairs; The quantitative data and the corresponding quantitative labels are used as quantitative label data pairs.

5. A method for quantitative text extraction in the real estate industry as described in claim 4, characterized in that, The quantitative label data pairs are validated, and the validation results are used to complete the quantitative label data pairs to obtain validated standard quantitative data pairs, including: Determine whether Chinese characters exist in the quantitative data of the quantitative label data pair; If Chinese characters are present in the quantitative data, the quantitative label data pair will be discarded. If no Chinese characters are present in the quantitative data, then determine whether there are unit symbols in the quantitative data. If the quantitative data does not contain a unit symbol, the quantitative data is supplemented with units according to a preset quantitative data template, and the supplemented quantitative label data pair is used as a standard quantitative data pair. If the quantitative data contains a unit of measurement symbol, then the quantitative data is determined to be within the preset numerical range based on the preset numerical range corresponding to the quantitative label. If the quantitative data does not conform to the preset numerical range, the quantitative label data pair will be discarded; If the quantitative data meets the preset numerical range, the quantitative label data pair is used as the standard quantitative data pair.

6. A quantitative text extraction system for the real estate industry, characterized in that, The system includes: The data acquisition module is used to acquire the dialogue text entered by real estate consultants after each visitor meeting; The text annotation module is used to perform tag matching on the dialogue text, annotate the dialogue text according to the tag matching results, and use the annotated dialogue text as the first tag text pair. The data cleaning module is used to clean the fuzzy quantifiers in the first tag text pair based on the tags in the first tag text pair, so as to obtain the second tag text pair. The quantitative text standardization module is used to standardize the quantitative text in the second label text pair according to the label type of the label in the second label text pair, so as to obtain a standard label text pair. The quantitative text extraction module is used to input the standard label text pairs into a pre-trained quantitative text extraction and annotation model to obtain quantitative label data pairs; The data verification module is used to verify the quantitative label data pairs, and to complete the quantitative label data pairs according to the verification results to obtain standard quantitative data pairs that have passed the verification. The dialogue text is subjected to tag matching, and the dialogue text is labeled according to the tag matching results. The labeled dialogue text is used as the first tag text, including: Based on the punctuation marks in the dialogue text, the dialogue text is segmented into short sentences to obtain the segmented short sentence text. For each of the short sentence texts, based on a preset word tag library, word tag matching is performed on the short sentence text to determine whether there is a word tag in the preset word tag library that matches the short sentence text; If no word tag in the preset word tag library matches the short sentence text, the short sentence text will be discarded. If there is a word tag in the preset word tag library that matches the short sentence text, then the short sentence text is annotated using the matched word tag, and the annotated short sentence text is used as the first annotated short sentence text. Using the preset word filter tag library corresponding to the word label, the first labeled short sentence text is matched with word filter tags to determine whether there are word filter tags in the preset word filter tag library that match the first labeled short sentence text. If no word filter tag in the preset word filter tag library matches the first annotated short sentence text, then the first annotated short sentence text is discarded; If there is a word filter tag in the preset word filter tag library that matches the first annotated short sentence text, then the first annotated short sentence text is annotated a second time using the matched word filter tag to obtain the second annotated short sentence text; The dialogue text corresponding to the second annotated short sentence text and the two adjacent dialogue texts before and after it are spliced ​​together in order to obtain the target long text; Using a preset sliding window, a preset number of short sentences are sequentially extracted from the beginning of the target long text to obtain the slid-extracted text; For the sliding text, calculate the vector similarity between the sliding text and each preset text in the preset corpus corresponding to the word filter tag, and obtain the vector similarity calculation result; Determine whether the vector similarity calculation result is greater than or equal to a preset similarity threshold; If the vector similarity calculation result is greater than or equal to the preset similarity threshold, then the first tag text pair is obtained based on the dialogue text corresponding to the second labeled short sentence text and the corresponding word label; If the vector similarity is less than a preset similarity threshold, then it is determined whether the sliding window has slid to the end of the target long text; If the sliding window does not slide to the end of the target long text, the sliding window slides at the current position to the end of the target long text by the distance of a short sentence text, and repeats until a preset number of short sentences text are extracted to obtain the slid-trimmed text; If the sliding window has slid to the end of the target long text, then the first tag text pair is obtained based on the second annotated short sentence text and the corresponding word filter tag; Based on the tags in the first tag text pair, the fuzzy quantifiers in the first tag text pair are cleaned to obtain the second tag text pair, including: The tags in the first tag text pair are matched with a preset first quantitative tag library to obtain the first matching result; Determine whether the first matching result is successful; If the first matching result fails, no data cleaning processing will be performed on the first tag text pair; If the first matching result is successful, then perform text structure detection on the text in the first tag text pair to obtain the detection result; Based on the detection results, determine whether there are fuzzy quantifiers in the text of the first tag text pair; If the text in the first tag text pair contains fuzzy quantifiers, then according to the preset fuzzy quantifier cleaning scheme, the fuzzy quantifiers are cleaned to obtain the second tag text pair; If the text in the first tag text pair does not contain any fuzzy quantifiers, then the first tag text pair is directly used as the second tag text pair.

7. A quantitative text extraction device for the real estate industry, characterized in that, The device includes: a processor and a memory; The memory is used to store one or more program instructions; The processor is configured to run one or more program instructions to perform the steps of a quantitative text extraction method for the real estate industry as described in any one of claims 1 to 5.

8. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the steps of a quantitative text extraction method for the real estate industry as described in any one of claims 1 to 5.

Images