A bidding and procurement demand analysis method and system based on text analysis

Abstract

The present invention discloses a method and system for analyzing bidding and procurement requirements based on text analysis. The method comprises the following steps: obtaining the text of a bidding and procurement announcement to be analyzed; extracting keywords from the text to be analyzed and obtaining keyword categories; classifying procurement requirements based on all keywords in the text to be analyzed; selecting at least one keyword from different categories of keywords and combining the selected keywords to form a procurement requirement analysis statement; and inputting the procurement requirement analysis statement into a preset procurement requirement analysis model to obtain procurement requirement features, wherein the procurement requirement analysis model employs an attention mechanism. The present invention analyzes the procurement requirement form based on the extraction of original material keywords and accurately extracts and reorganizes them according to the specific attributes of the materials, thereby mining the true procurement requirements of the bidding and procurement.

Description

Technical Field

[0001] The present invention relates to the technical field of tendering and procurement analysis, and in particular to a tendering and procurement demand analysis method and system based on text analysis. Background Art

[0002] Every day, tens of thousands of bidding and tender information and winning bid announcements are published on government portals, government procurement websites, industry bidding platforms, and other media nationwide. This information covers a wide range of topics, including a wide variety of procurement needs. For procurement platforms, timely accessing this vast amount of information, identifying the real procurement needs, and processing them accordingly is an extremely valuable yet challenging task. Generally, this information includes key elements such as the bidding company, tender type, tender product, contact number, procurement quantity, bid results, project name, deadline, and qualification requirements. This information is presented in the form of unstructured or semi-structured text within the platform's tender announcements. Analyzing and identifying these elements to identify the real procurement needs and corresponding services is crucial for the procurement platform's accurate matching and recommendation of suppliers, and is a crucial market task.

[0003] In the existing technology, simply structuring the attributes of required materials has the problems of a wide range of procurement requirements, inaccurate materials, and unclear demand identification. Summary of the Invention

[0004] To address the above-mentioned problems in the existing technology, the present invention provides a method and system for analyzing bidding and procurement requirements based on text analysis, which effectively improves the accuracy of procurement requirements analysis. The technical solution is as follows:

[0005] First, a method for analyzing bidding and procurement requirements based on text analysis is provided, including:

[0006] Obtain the text of the tender and procurement notice to be analyzed, and record it as the text to be analyzed;

[0007] Extracting keywords from the text to be analyzed and obtaining categories of the keywords; the categories of the keywords are determined based on a plurality of preset categories; the plurality of preset categories include at least one category representing a material feature category, at least one category representing a material attribute category, and at least one category representing a material service category;

[0008] Classifying procurement demand forms based on all keywords in the text to be analyzed, wherein the procurement demand forms include single form and multiple form;

[0009] Selecting at least one keyword from different categories of keywords and combining the selected keywords to form a procurement demand analysis statement, wherein when the procurement demand form is single-state, the categories of the selected keywords include at least a material feature category; and when the procurement demand form is multi-state, the categories of the selected keywords include at least a material feature category and a material service category;

[0010] The procurement demand analysis statement is input into a preset procurement demand analysis model to obtain procurement demand features, and the procurement demand analysis model adopts an attention mechanism.

[0011] In some implementations, extracting keywords from the text to be analyzed and performing semantic recognition and classification on the keywords includes:

[0012] Obtaining a training text, wherein the training text is a text of a bidding and procurement announcement;

[0013] Based on the training text, each character is matched in a preset dictionary to obtain at least one matching word;

[0014] Get the position information of each character in each matching word;

[0015] Get the corresponding word feature vector for each matching word of the character;

[0016] Perform weighted fusion on all word feature vectors of a character by matching the preset weight of the word to obtain the fused word feature vector of the character;

[0017] Obtaining a feature vector of a character, and concatenating the feature vector of the character with a fusion word feature vector of the character as a character representation feature, inputting all the character representation features in the training text into a preset first neural network, training the preset first neural network with the position of preset category keywords corresponding to the training text and keyword category annotation data, and using the trained preset first neural network as a keyword recognition model;

[0018] For the text to be analyzed, a preset dictionary is used to obtain the character representation features of each character in the text to be analyzed;

[0019] All character representation features of the text to be analyzed are input into the keyword recognition model to obtain the keywords and keyword categories of the text to be analyzed.

[0020] In some implementations, the method for obtaining the word feature vector of the matching word is:

[0021] The words in the preset dictionary are divided into small categories according to synonyms or near-synonyms, and multiple small categories are merged into a large category according to the semantic categories of the words;

[0022] Perform one-hot encoding on each major category to obtain a first code value, perform one-hot encoding on each subcategory in each major category in turn to obtain a second code value, perform one-hot encoding on each word in each subcategory in turn to obtain a third code value, and determine the code value of each word based on the splicing of the first code value, the second code value, and the third code value;

[0023] Inputting the encoding value of the word into a preset second neural network, training the second neural network with the encoding value of the context word of the word as the target output, and using the weight parameters of the trained second neural network as the word feature vector of the word;

[0024] The method for obtaining the feature vector of the character is:

[0025] The feature vector of each character is determined based on the number of matching words of the character, the number of occurrences of the matching words, and the word feature vector of each matching word.

[0026] In some embodiments, the preset multiple categories of keywords include: project, equipment, model, manufacturer, quantity and service type, and project includes supplies.

[0027] In some embodiments, selecting at least one keyword from different categories of keywords and combining the selected keywords to form a procurement demand analysis statement includes:

[0028] Based on at least one keyword of at least one selected category, the keywords are combined in the order of material attribute class, material feature class, and material service class to form a procurement demand analysis statement.

[0029] In some embodiments, inputting the procurement demand analysis statement into a preset procurement demand analysis model to obtain procurement demand characteristics includes:

[0030] Encode all keywords in the procurement demand analysis statement to obtain word vectors and form a word vector sequence;

[0031] Obtain dependency features between word vectors in a word vector sequence through an attention mechanism, fuse the dependency features into the word vector sequence, and obtain a fused word vector sequence;

[0032] Based on the fused word vector sequence, the third neural network in the procurement demand analysis model is used to obtain the procurement demand feature representation.

[0033] In some embodiments, obtaining dependency features between word vectors in a word vector sequence through an attention mechanism includes:

[0034] Get the position distance between any word vector and the remaining word vectors in the word vector sequence;

[0035] Obtain the correlation between any word vector and the remaining word vectors in the word vector sequence, where the correlation represents the synchronization of the two word vectors.

[0036] Determine a first relevance feature by determining a word vector based on the position distance and the relevance;

[0037] Determining a second relevance feature based on a difference between any word vector and the remaining word vectors in the word vector sequence;

[0038] Obtaining a correlation matrix based on the first correlation feature and the second correlation feature;

[0039] Use the Softmax function to process the correlation matrix and the transpose of the correlation matrix respectively to obtain the attention score of any word vector to the remaining word vectors in the word vector sequence and the attention score of the remaining word vectors in the word vector sequence to any word vector;

[0040] Based on the attention scores of the remaining word vectors in the word vector sequence, the arbitrary word vector is corrected to obtain the first dependency feature. Based on the attention scores of the remaining word vectors in the word vector sequence, the remaining word vectors in the word vector sequence are corrected to obtain the second dependency feature. The first dependency feature and the second dependency feature are spliced ​​to obtain the dependency feature.

[0041] Secondly, a bidding and procurement demand analysis system based on text analysis is provided, including:

[0042] A text-to-be-analyzed acquisition unit is used to acquire the text of the tendering and procurement notice to be analyzed, which is recorded as the text to be analyzed;

[0043] A keyword extraction and demand form classification unit is used to extract keywords from the text to be analyzed and obtain keyword categories; the keyword categories are determined based on multiple preset categories; the multiple preset categories include at least one category representing material characteristics, at least one category representing material attributes, and at least one category representing material services; based on all keywords in the text to be analyzed, the procurement demand forms are classified into single-state and multi-state categories;

[0044] a procurement demand analysis statement acquisition unit, configured to select at least one keyword from among keywords of different categories and combine the selected keywords to form a procurement demand analysis statement; when the procurement demand form is single-state, the categories of the selected keywords include at least a material feature category; when the procurement demand form is multi-state, the categories of the selected keywords include at least a material feature category and a material service category;

[0045] The procurement demand feature acquisition unit is used to input the procurement demand analysis statement into a preset procurement demand analysis model to obtain procurement demand features, and the procurement demand analysis model adopts an attention mechanism.

[0046] In a third aspect, a computer-readable storage medium is provided, on which computer instructions are stored, characterized in that when the instructions are executed by a processor, the steps of the bidding and procurement demand analysis method based on text analysis as described in the first aspect above are implemented.

[0047] The present invention provides a method and system for analyzing bidding and procurement requirements based on text analysis, which has the following beneficial effects: analyzing the form of procurement requirements based on the extraction of original material keywords, and accurately extracting and reorganizing them according to the specific attributes of the materials, thereby improving the accuracy of procurement requirement mining in bidding and procurement. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 This is a schematic diagram of the overall process of the bidding and procurement demand analysis method according to the embodiment of the present application;

[0049] Figure 2 This is a flowchart of a method for analyzing keywords in a bidding and procurement announcement text according to an embodiment of the present application;

[0050] Figure 3 This is a schematic diagram of the attention mechanism for analyzing procurement demand analysis statements in an embodiment of the present application;

[0051] Figure 4 It is a structural diagram of the bidding and procurement demand analysis system of the embodiment of this application. DETAILED DESCRIPTION

[0052] The technical solution of the present invention is further described below.

[0053] See also Figure 1 The embodiment of the present application provides a method for analyzing bidding and procurement requirements based on text analysis, which is characterized by comprising the following steps:

[0054] Step 1: Obtain the text of the tender and procurement announcement to be analyzed, which is recorded as the text to be analyzed. The text to be analyzed includes the title, abstract and full text of the announcement;

[0055] Step 2: extract keywords from the text to be analyzed and obtain keyword categories; the keyword categories are determined based on a plurality of preset categories; the plurality of preset categories include at least one category representing a material feature category, at least one category representing a material attribute category, and at least one category representing a material service category;

[0056] Step 3: Classify the procurement demand forms based on all keywords in the text to be analyzed, where the procurement demand forms include single form and multiple form.

[0057] Step 4: Select at least one keyword from the different categories of keywords and combine the selected keywords to form a procurement demand analysis statement. When the procurement demand form is single-state, the categories of the selected keywords include at least material feature categories. When the procurement demand form is multi-state, the categories of the selected keywords include at least material feature categories and material service categories.

[0058] Step 5: Input the procurement demand analysis statement into a preset procurement demand analysis model to obtain procurement demand features, wherein the procurement demand analysis model adopts an attention mechanism.

[0059] Then, step 6 can be executed to match buyers and suppliers based on the procurement demand characteristics to form recommendation information corresponding to the bidding procurement demand.

[0060] Taking into account that the real demand may refer to the purchase of materials, but it may also be other services for existing materials, in the embodiment of the present application, the form of procurement demand is analyzed based on the extraction of original material keywords, and accurate extraction and reorganization are performed according to the specific attributes of the materials, so as to explore the real procurement needs of bidding procurement.

[0061] For further information, see Figure 2 In the above step 2, extracting keywords from the text to be analyzed and performing semantic recognition and classification on the keywords include:

[0062] Step 21: obtaining a training text, wherein the training text is the text of a bidding and procurement announcement;

[0063] Step 22: Based on the training text, match each character in a preset dictionary to obtain at least one matching word;

[0064] Step 23, obtaining the position information of each character in each matching word;

[0065] Step 24, obtaining the corresponding word feature vector for each matching word of the character;

[0066] Step 25, performing weighted fusion on all word feature vectors of the character using the preset weight of the matching word to obtain the fused word feature vector of the character;

[0067] Step 26: Obtain a character feature vector, and concatenate the character feature vector and the character fusion word feature vector as a character representation feature, input all character representation features in the training text into a preset first neural network, train the preset first neural network using the positions of preset category keywords corresponding to the training text and keyword category annotation data, and use the trained preset first neural network as a keyword recognition model;

[0068] Step 27: For the text to be analyzed, a preset dictionary is used to obtain the character representation features of each character in the text to be analyzed;

[0069] In step 28, all character representation features of the text to be analyzed are input into a keyword recognition model to obtain keywords and keyword categories of the text to be analyzed.

[0070] In an embodiment of the present application, when extracting keywords, all possible matching words are extracted for each character, taking into account the problem of the uncertain position of each character in the composed words, and all matching words corresponding to the character at different positions in the word are obtained. Furthermore, the position of the character with greater possibility in the word is judged according to the feature vectors and preset weights of different matching words, thereby enhancing the accuracy of keyword position analysis; at the same time, the present application combines the feature vector of the character itself and the feature vector of the fusion word of the character, and uses them as character representation features to perform keyword extraction and keyword classification analysis of the text, taking into account the semantic information of the characters in the text environment during the keyword analysis process, thereby improving the accuracy of keyword analysis.

[0071] Furthermore, in the above step 24, the method for obtaining the word feature vector of the matching word is:

[0072] Step 241: Classify the words in the preset dictionary into subcategories according to synonyms or near-synonyms, and merge multiple subcategories into a larger category according to the semantic categories of the words;

[0073] Step 242: Perform one-hot encoding on each major category to obtain a first encoding value, perform one-hot encoding on each subcategory in each major category to obtain a second encoding value, perform one-hot encoding on each word in each subcategory to obtain a third encoding value, and determine the encoding value of each word based on the concatenation of the first encoding value, the second encoding value, and the third encoding value.

[0074] Step 243: Input the encoding value of the word into a preset second neural network, train the second neural network with the encoding value of the context word of the word as the target output, and use the weight parameters of the trained second neural network as the word feature vector of the word;

[0075] In the above step 26, the method for obtaining the feature vector of the character is:

[0076] The feature vector of each character is determined based on the number of matching words of the character, the number of occurrences of the matching words, and the word feature vector of each matching word.

[0077] In the embodiment of the present application, the classification of small categories and large categories makes the word encoding more able to represent the close distance between similar words and the long distance characteristics between irrelevant words; at the same time, in the embodiment of the present application, the first encoding value of the large category encoding, the second encoding value of the small category encoding, and the splicing of the first encoding value, the second encoding value, and the third encoding value are used to determine the encoding value of each word. On the one hand, it ensures that the encoding value effectively represents the similarity of the words. On the other hand, based on the effective reduction of the encoding length of directly encoding each word based on the large category encoding and the small category encoding, the efficiency of the word feature vector obtained by the second neural network is effectively realized. Among them, in the method for obtaining the feature vector of the character, the word feature vector of the matching word can be weighted based on the number of matching words of the character and the number of occurrences of the matching word to obtain the feature vector of the character.

[0078] Furthermore, in step 2 above, the preset multiple categories of keywords include: project, equipment, model, manufacturer, quantity, and service type, where projects include supplies. Project / supplies and equipment can be used as keywords for material characteristics, model, manufacturer, and quantity can be used as keywords for material attributes, and service type can be used as keywords for material services. For example, the material characteristic word is "wardrobe," and the material attributes are "Manufacturer: OPPEIN," "Model: Simple European," "Quantity: 32 square meters," and "Service Type: Whole House."

[0079] Furthermore, in the above step 3, the procurement demand forms are classified based on all the keywords of the text to be analyzed. All the keywords obtained in step 2 can be used as inputs of a preset third neural network model, with the category of the procurement demand form as a single state or a multiple state as output. The third neural network is trained with all the keywords of the sample tender procurement text as input and the category of the procurement demand form corresponding to the sample tender procurement text as the target output. In another embodiment, the category classification of the procurement demand form can be obtained together with the category of the keywords in step 26, that is, in step 26, all the character representation features in the training text are input into the preset first neural network, and the position of the preset category keyword corresponding to the training text, the keyword category and the category of the procurement demand form corresponding to the training text are used as three kinds of sample annotation data to train the preset first neural network, and the preset first neural network after training is used as a keyword recognition model; the keyword recognition model can simultaneously recognize the keyword position, keyword category and the category of the procurement demand form corresponding to the tender procurement text.

[0080] Furthermore, in the above step 4, at least one keyword of different categories is selected, and the selected keywords are combined to form a procurement demand analysis statement, including:

[0081] Step 41 : Based on at least one keyword of at least one selected category, the keywords are combined in the order of material attribute class, material feature class, and material service class to form a procurement demand analysis statement.

[0082] Furthermore, when the category of the procurement demand form is single-state, the keyword category combination can be material feature keywords; material attribute keywords + material feature keywords; when the category of the procurement demand form is polymorphic, the keyword category combination can be material feature + material service; material attribute + material feature + material service;

[0083] For example, the keywords of the bidding and procurement announcement text are: the demand service trigger word is "customization", the material characteristic word is "wardrobe", and the material attributes are "manufacturer: Oppein", "model: simple European style", "quantity: 32 square meters", and "service type: whole house"; then the procurement demand analysis statement can be: "Wardrobe whole house customization", "Oppein wardrobe whole house customization", "Simple European Oppein wardrobe whole house customization", "32 square meters simple European Oppein wardrobe whole house customization", etc.

[0084] Furthermore, in the above step 5, the procurement demand analysis statement is input into the preset procurement demand analysis model to obtain procurement demand characteristics, including:

[0085] Step 51: Encode all keywords in the procurement demand analysis statement to obtain word vectors to form a word vector sequence;

[0086] Step 52: Obtain dependency features between word vectors in the word vector sequence through an attention mechanism, fuse the dependency features into the word vector sequence, and obtain a fused word vector sequence;

[0087] Step 53: Obtain procurement demand feature representation using the third neural network in the procurement demand analysis model based on the fused word vector sequence.

[0088] In an embodiment of the present application, when analyzing the procurement requirement characteristics of the tender procurement announcement text, the dependency characteristics between all keywords in the procurement requirement analysis statement are taken into consideration, thereby effectively improving the effective information expression of the procurement requirement characteristics obtained by the procurement requirement analysis model and achieving the accuracy of the procurement requirement analysis.

[0089] For further information, see Figure 3 The above 52 steps, which use the attention mechanism to obtain the dependency features between word vectors in the word vector sequence, include:

[0090] Step 521, obtaining the position distance between any word vector and the remaining word vectors in the word vector sequence;

[0091] Step 522: Obtain the correlation between any word vector and the remaining word vectors in the word vector sequence, where the correlation represents the synchronization of the appearance of the two word vectors.

[0092] Step 523: Determine a first relevance feature by determining a word vector based on the position distance and the relevance;

[0093] Step 524: determining a second relevance feature based on the difference between the arbitrary word vector and the remaining word vectors in the word vector sequence;

[0094] Step 525 , obtaining a correlation matrix based on the first correlation feature and the second correlation feature;

[0095] Step 526: Use the Softmax function to process the correlation matrix and the transpose of the correlation matrix to obtain the attention score of any word vector to the remaining word vectors in the word vector sequence and the attention score of the remaining word vectors in the word vector sequence to any word vector;

[0096] Step 527: Based on the attention scores of the remaining word vectors in the word vector sequence, any word vector is corrected to obtain a first dependency feature; based on the attention scores of the remaining word vectors in the word vector sequence, any word vector is corrected to obtain a second dependency feature; the first dependency feature and the second dependency feature are concatenated to obtain a dependency feature.

[0097] In an embodiment of the present application, the attention scores between word vectors are obtained through the Softmax function based on the correlation matrix and the transpose of the correlation matrix, and each word vector is processed based on the attention score to obtain the attention feature representation, that is, the dependency feature, and further based on the attention feature representation and the original word vector combination to form a fused word vector sequence, and the fused word vector sequence is used to obtain the procurement demand feature representation using the second neural network in the procurement demand analysis model. It can be understood that the attention feature representation includes the multi-faceted correlation relationship features of the position distance of the word, the correlation of the word vector, and the difference of the word vector, so as to better mine the global feature of all keywords in the procurement demand analysis statement. Among them, the difference in word vectors in step 524 can be calculated based on the distance between the word vectors.

[0098] See also Figure 4 The embodiment of the present application further provides a bidding and procurement demand analysis system based on text analysis, the system comprising:

[0099] A text-to-be-analyzed acquisition unit is used to acquire the text of the tendering and procurement notice to be analyzed, which is recorded as the text to be analyzed;

[0100] A keyword extraction and demand form classification unit is used to extract keywords from the text to be analyzed and obtain keyword categories; the keyword categories are determined based on multiple preset categories; the multiple preset categories include at least one category representing material characteristics, at least one category representing material attributes, and at least one category representing material services; based on all keywords in the text to be analyzed, the procurement demand forms are classified into single-state and multi-state categories;

[0101] a procurement demand analysis statement acquisition unit, configured to select at least one keyword from among keywords of different categories and combine the selected keywords to form a procurement demand analysis statement; when the procurement demand form is single-state, the categories of the selected keywords include at least a material feature category; when the procurement demand form is multi-state, the categories of the selected keywords include at least a material feature category and a material service category;

[0102] The procurement demand feature acquisition unit is used to input the procurement demand analysis statement into a preset procurement demand analysis model to obtain procurement demand features, and the procurement demand analysis model adopts an attention mechanism.

[0103] It should be noted that the text analysis-based bidding and procurement demand analysis system provided in this embodiment only uses the division of the above-mentioned functional units as an example when analyzing the procurement requirements of the bidding and procurement announcement text. In actual applications, the above-mentioned functions can be assigned to different functional units as needed, that is, the internal structure of the device can be divided into different functional units to complete all or part of the functions described above. In addition, the bidding and procurement demand analysis system provided in this embodiment and the bidding and procurement demand analysis method embodiment provided in the above embodiment belong to the same concept. The specific implementation process is detailed in the method embodiment and will not be repeated here.

[0104] The embodiment of the present application also provides a computer-readable storage medium having computer instructions stored thereon, characterized in that when the instructions are executed by a processor, the steps of the above-mentioned bidding and procurement demand analysis method based on text analysis are implemented. It can be understood that computer-readable storage media include: permanent and non-permanent, removable and non-removable media, which are tangible devices that can retain and store instructions for use by instruction execution devices. For example, the computer-readable storage medium can be a read-only memory (ROM), a random access memory (RAM), a compact disc read-only memory (CD-ROM), etc.

[0105] The present invention is not limited to the above-mentioned specific implementation methods. Various changes made by ordinary technicians in this field based on the above-mentioned concept without creative work are all within the scope of protection of the present invention.

Claims

1. A method for analyzing bidding and procurement requirements based on text analysis, characterized in that: include: Obtain the text of the tender and procurement notice to be analyzed, and record it as the text to be analyzed; Extract keywords from the text to be analyzed and obtain the categories of the keywords; The category of the keyword is determined based on a plurality of preset categories; the plurality of preset categories include at least one category representing a material feature category, at least one category representing a material attribute category, and at least one category representing a material service category; Classifying procurement demand forms based on all keywords in the text to be analyzed, wherein the procurement demand forms include single form and multiple form; Selecting at least one keyword from different categories of keywords and combining the selected keywords to form a procurement demand analysis statement, wherein when the procurement demand form is single-state, the categories of the selected keywords include at least a material feature category; and when the procurement demand form is multi-state, the categories of the selected keywords include at least a material feature category and a material service category; Inputting the procurement demand analysis statement into a preset procurement demand analysis model to obtain procurement demand features, wherein the procurement demand analysis model adopts an attention mechanism; The method of extracting keywords from a text to be analyzed and obtaining the categories of keywords includes: obtaining a training text, which is a text of a bidding and procurement announcement; matching each character in a preset dictionary based on the training text to obtain at least one matching word; obtaining the position information of each character in each matching word; obtaining a corresponding word feature vector for each matching word of the character; weightedly fusing all word feature vectors of the character by using a preset weight of the matching word to obtain a fused word feature vector of the character; obtaining a feature vector of the character, and connecting the feature vector of the character with the fused word feature vector of the character as a character representation feature, inputting all the character representation features in the training text into a preset first neural network, training the preset first neural network with the position of the preset category keyword corresponding to the training text and the keyword category annotation data, and using the trained preset first neural network as a keyword recognition model; for the text to be analyzed, obtaining the character representation features of each character in the text to be analyzed using a preset dictionary; inputting all the character representation features of the text to be analyzed into the keyword recognition model to obtain the keywords and keyword categories of the text to be analyzed; The method of inputting the procurement demand analysis statement into the preset procurement demand analysis model to obtain procurement demand features includes: encoding all keywords in the procurement demand analysis statement to obtain word vectors to form a word vector sequence; obtaining dependency features between word vectors in the word vector sequence through an attention mechanism, fusing the dependency features into the word vector sequence to obtain a fused word vector sequence; and obtaining procurement demand feature representation based on the fused word vector sequence using the third neural network in the procurement demand analysis model.

2. The method for analyzing bidding and procurement requirements based on text analysis according to claim 1, characterized in that: The method for obtaining the word feature vector of the matching word is: The words in the preset dictionary are divided into small categories according to synonyms or near-synonyms, and multiple small categories are merged into a large category according to the semantic categories of the words; Perform one-hot encoding on each major category to obtain a first code value, perform one-hot encoding on each subcategory in each major category in turn to obtain a second code value, perform one-hot encoding on each word in each subcategory in turn to obtain a third code value, and determine the code value of each word based on the splicing of the first code value, the second code value, and the third code value; Inputting the encoding value of the word into a preset second neural network, training the second neural network with the encoding value of the context word of the word as the target output, and using the weight parameters of the trained second neural network as the word feature vector of the word; The method for obtaining the feature vector of the character is: The feature vector of each character is determined based on the number of matching words of the character, the number of occurrences of the matching words, and the word feature vector of each matching word.

3. The method for analyzing bidding and procurement requirements based on text analysis according to claim 1, characterized in that: The preset multiple categories of keywords include: project, equipment, model, manufacturer, quantity and service type, and projects include materials.

4. The method for analyzing bidding and procurement requirements based on text analysis according to claim 1, characterized in that: The step of selecting at least one keyword from the different categories of keywords and combining the selected keywords to form a procurement demand analysis statement includes: Based on at least one keyword of at least one selected category, the keywords are combined in the order of material attribute class, material feature class, and material service class to form a procurement demand analysis statement.

5. The method for analyzing bidding and procurement requirements based on text analysis according to claim 1, characterized in that: The dependency features between word vectors in a word vector sequence are obtained through the attention mechanism, including: Get the position distance between any word vector and the remaining word vectors in the word vector sequence; Obtain the correlation between any word vector and the remaining word vectors in the word vector sequence, where the correlation represents the synchronization of the two word vectors. Determine a first relevance feature by determining a word vector based on the position distance and the relevance; Determining a second relevance feature based on a difference between any word vector and the remaining word vectors in the word vector sequence; Obtaining a correlation matrix based on the first correlation feature and the second correlation feature; Use the Softmax function to process the correlation matrix and the transpose of the correlation matrix respectively to obtain the attention score of any word vector to the remaining word vectors in the word vector sequence and the attention score of the remaining word vectors in the word vector sequence to any word vector; Based on the attention scores of the remaining word vectors in the word vector sequence, the arbitrary word vector is corrected to obtain the first dependency feature. Based on the attention scores of the remaining word vectors in the word vector sequence, the remaining word vectors in the word vector sequence are corrected to obtain the second dependency feature. The first dependency feature and the second dependency feature are spliced ​​to obtain the dependency feature.

6. A bidding and procurement demand analysis system based on text analysis, characterized in that: include: A text-to-be-analyzed acquisition unit is used to acquire the text of the tendering and procurement notice to be analyzed, which is recorded as the text to be analyzed; A keyword extraction and demand form classification unit is used to extract keywords from the text to be analyzed and obtain keyword categories; the keyword categories are determined based on multiple preset categories; the multiple preset categories include at least one category representing material characteristics, at least one category representing material attributes, and at least one category representing material services; based on all keywords in the text to be analyzed, the procurement demand forms are classified into single-state and multi-state categories; a procurement demand analysis statement acquisition unit, configured to select at least one keyword from among keywords of different categories and combine the selected keywords to form a procurement demand analysis statement; when the procurement demand form is single-state, the categories of the selected keywords include at least a material feature category; when the procurement demand form is multi-state, the categories of the selected keywords include at least a material feature category and a material service category; a procurement demand feature acquisition unit, configured to input a procurement demand analysis statement into a preset procurement demand analysis model to acquire procurement demand features, wherein the procurement demand analysis model adopts an attention mechanism; In the keyword extraction and demand morphology classification unit, keywords of the text to be analyzed are extracted, and the categories of keywords are obtained, including: obtaining a training text, which is the text of a bidding and procurement announcement; based on the training text, matching each character in a preset dictionary to obtain at least one matching word; obtaining the position information of each character in each matching word; obtaining the corresponding word feature vector for each matching word of the character; weighted fusion of all word feature vectors of the character by the preset weight of the matching word to obtain the fused word feature vector of the character; obtaining the feature vector of the character, and connecting the feature vector of the character and the fused word feature vector of the character as a character representation feature, inputting all the character representation features in the training text into a preset first neural network, training the preset first neural network with the position of the preset category keyword corresponding to the training text and the keyword category annotation data, and using the trained preset first neural network as a keyword recognition model; for the text to be analyzed, obtaining the character representation features of each character in the text to be analyzed using a preset dictionary; inputting all the character representation features of the text to be analyzed into the keyword recognition model to obtain the keywords and keyword categories of the text to be analyzed; In the procurement demand feature acquisition unit, the procurement demand analysis statement is input into the preset procurement demand analysis model to obtain procurement demand features, including: encoding all keywords in the procurement demand analysis statement to obtain word vectors to form a word vector sequence; obtaining the dependency features between word vectors in the word vector sequence through the attention mechanism, and fusing the dependency features into the word vector sequence to obtain a fused word vector sequence; and obtaining the procurement demand feature representation based on the fused word vector sequence using the third neural network in the procurement demand analysis model.

7. A computer-readable storage medium having computer instructions stored thereon, characterized in that: When the instructions are executed by the processor, the steps of the bidding and procurement demand analysis method based on text analysis as described in any one of claims 1 to 5 are implemented.

Images