Document management system based on data analysis

Abstract

The invention relates to the technical field of literature management, in particular to a literature management system based on data analysis. The system comprises the steps that document information is obtained, and the document information comprises document titles; a mark sequence is predicted and written for the literature title; predicting a part-of-speech tagging sequence for the literature title: converting a sentence corresponding to the literature title into an input vector through a FastText word embedding technology, inputting the input vector into the bidirectional LSTM network, and outputting the part-of-speech tagging sequence; performing named entity labeling on the literature titles; and carrying out document classification on the basis of the document titles which contain the division-writing tagging sequences and the part-of-speech tagging sequences and are tagged by the named entities. According to the method, the literature can be automatically classified, labeled and retrieved based on the title of the literature, the intelligent processing function of the literature management system is improved, accurate classification is realized, and the literature management efficiency is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of document management, and in particular to a document management system based on data analysis. Background Art

[0002] Document management refers to a series of activities such as the collection, organization, analysis, storage and citation of academic documents. It is an indispensable part of the academic research process.

[0003] Traditional document management methods rely on manual classification, annotation and retrieval, which is inadequate when faced with massive amounts of document data. The rapid growth of document information makes it difficult for researchers and students to efficiently collect, organize, analyze and utilize these resources. At the same time, traditional document management systems often lack intelligent processing functions, cannot automatically extract key information from documents, cannot achieve accurate classification, and further cannot make accurate recommendations based on users' personalized needs.

[0004] Therefore, there is an urgent need for a document management system based on data analysis to solve the above problems. Summary of the invention

[0005] The purpose of the present invention is to provide a document management system based on data analysis: it solves the technical problem that traditional document management methods rely heavily on manual classification, labeling and retrieval, resulting in document management systems often lacking intelligent processing functions, being unable to automatically extract key information from documents, and being unable to achieve accurate classification.

[0006] The purpose of the present invention can be achieved through the following technical solutions:

[0007] A document management system based on data analysis, the system includes a document information acquisition module, a document title automatic segmentation module, a document title part-of-speech tagging module, a document named entity recognition module and a document classification module;

[0008] The document information acquisition module is used to acquire document information, wherein the document information includes the document title;

[0009] The automatic segmentation module of the document title is used to predict the segmentation annotation sequence of the document title: convert the sentence corresponding to the document title into X=(x 1 , x 2 ,...,x n ), x i Represents the index value of the i-th character, i = 1, 2...n, input X into the character query table, and query each x i Convert it into a low-dimensional real number vector of fixed length, combine the low-dimensional real number vector into the input sentence LX(X), and input LX(X) into the bidirectional LSTM network to output the segmented annotation sequence;

[0010] The document title part-of-speech tagging module is used to predict the part-of-speech tagging sequence for the document title: the sentence corresponding to the document title is converted into an input vector using the FastText word embedding technology, and the input vector is input into the bidirectional LSTM network to output the part-of-speech tagging sequence;

[0011] The document named entity recognition module is used to annotate document titles with named entities;

[0012] The document classification module is used to classify documents based on document titles that contain a subscript tag sequence and a part-of-speech tag sequence and are annotated with named entities.

[0013] Further, input X into the character query table, and query each x i The conversion into a low-dimensional real vector of fixed length specifically includes the following processes:

[0014] Through FastText word embedding technology, each x is transformed into i Convert to a low-dimensional real vector of fixed length:

[0015] The character sequence that constitutes the word is split into a set of n-grams, and the original word is added to form the basic input unit of model training. In the fastText model, the sum of the vectors of the subsequences in the n-gram set represents a word. For each x i Converted into a low-dimensional real vector u of fixed length t :

[0016]

[0017] Among them, Z j Represents x i The n-gram vector is obtained by table lookup of the word vector matrix, j=1,2...m, m is the number of n-gram vectors obtained by table lookup of the word vector matrix.

[0018] Furthermore, LX(X) is input into the bidirectional LSTM network, and the output of the segmented label sequence specifically includes the following process:

[0019] The CRF layer of the LSTM network predicts the segmented annotation sequence:

[0020] S char (i) = f(W out hi+b out );

[0021]

[0022] Among them, S char(i) represents x i The probability distribution of the split annotation obtained by the bidirectional LSTM network, W out 、b out are the mapping matrix and bias vector of the fully connected layer, f is the softmax function, A zy The transition matrix of the split-write annotation state, S(X, Y 1 , θ) represents a candidate tag sequence Y corresponding to the input character sequence X when considering the tag state transition probability and the split tag probability predicted by the bidirectional LSTM 1 The score of , where θ represents the model parameters;

[0023] Take the score S(X, Y) from all candidate annotation paths 1 ,θ) is the largest path as the segmentation annotation sequence.

[0024] Furthermore, the input vector is input into the bidirectional LSTM network, and the output part-of-speech tag sequence includes the following process:

[0025] Predict the part-of-speech tag sequence through the CRF layer of the LSTM network:

[0026] S sylla (i) = f(W out h i +b out );

[0027]

[0028] Among them, S sylla (i) represents the probability distribution of part-of-speech tags obtained by the input vector through the bidirectional LSTM network, W out 、b out are the mapping matrix and bias vector of the fully connected layer, f is the softmax function, A zx represents the transfer matrix of part-of-speech tagging, S(X, Y 2 ,δ) represents a candidate tag sequence Y corresponding to the input vector when considering the tag state transition probability and the probability of writing tags predicted by the bidirectional LSTM 2 The score of , where δ represents the model parameter;

[0029] Take the score S(X, Y) from all candidate annotation paths 2 ,δ) is the largest path as the part-of-speech tag sequence.

[0030] Furthermore, named entity annotation of document titles specifically includes the following processes:

[0031] Given a document title input sentence w = (w 1 , w 2 ,...,wn ), w i represents the i-th morpheme in the input sentence w of the document title, in, Represents the morphogen w i The gth syllable of ;

[0032] The document title is input into sentence w and processed by FastText word embedding technology to obtain the syllables represented by low-dimensional real number vectors The vector representing the g-th syllable in the i-th morpheme;

[0033] Will Input into the bidirectional LSTM network to get the last state of the forward LSTM network And the last state of the backward LSTM network

[0034]

[0035] and Represent the forward state and backward state operations respectively;

[0036] Will Concatenate into a vector γ i :

[0037] Among them, * represents the concatenation operation;

[0038] Will γ i Input into the fully connected network and output the named entity label of the morpheme through a softmax function.

[0039] Furthermore, γ i The input is fed into a fully connected network, and the named entity label of the morpheme is output through a softmax function. The specific process includes the following:

[0040] p v (t) = softmax(W x γ i +b x );

[0041] p v (t) represents the probability distribution of named entity labels predicted using the syllable sequence in the morpheme, W x , b x represents trainable parameters;

[0042] The objective function is:

[0043] Among them, t lkrepresents the kth named entity label of the lth morpheme. If the named entity of the lth morpheme belongs to the kth named entity category, then t lk is 1, otherwise, t lk is 0, represents the probability that the lth morpheme belongs to the kth named entity category, N is the number of morphemes, and C is the number of named entity categories.

[0044] Furthermore, the document classification based on the document title containing the subscript tag sequence and the part-of-speech tag sequence and annotated with the named entity specifically includes the following process:

[0045] Step 1: parse the annotation information:

[0046] Separate notations distinguish different clauses or phrases in a title;

[0047] Part-of-speech tagging provides the grammatical role each word plays in a sentence;

[0048] Named entity annotation points out important entities in the text;

[0049] Step 2: Extract key information:

[0050] Use named entity annotation to directly identify key entities in the title;

[0051] Combined with part-of-speech tagging, the verb phrases and noun phrases in the title are identified;

[0052] Step 3: Build a classification system:

[0053] Construct a classification system based on the research field, research type, and application object to which the literature belongs;

[0054] Step 4: Document classification:

[0055] Match the parsed and extracted key information with the classification system to determine the category to which the document title belongs.

[0056] Furthermore, matching the parsed and extracted key information with the classification system to determine the category to which the document title belongs specifically includes the following steps:

[0057] Keyword matching: Match the extracted keywords with the various category labels in the classification system; the matching process is based on the contextual information of the keywords in the title and their relationship with other words in the title;

[0058] Weight allocation: During the matching process, different key information is assigned different weights;

[0059] Determine the category: Determine the category to which the document title belongs based on the matching results of key information and the classification system.

[0060] Compared with the existing solutions, the present invention achieves the following beneficial effects:

[0061] On the one hand, the present invention can classify, annotate and retrieve documents based on their titles, thereby improving the intelligent processing function of the document management system, achieving accurate classification and improving document management efficiency.

[0062] On the other hand, the present invention improves the efficiency of document retrieval, optimizes document organization and classification, supports in-depth analysis and mining, and improves the efficiency of document citation and writing.

[0063] Thirdly, the present invention can realize the functions of online collaboration and social sharing. Team members can upload data analysis attachments anytime and anywhere, and multiple people can edit them in real time, which reduces version conflicts and speeds up the data analysis process, thus improving the efficiency of document management and the dissemination of results.

[0064] Fourthly, the present invention can realize the function of hierarchical authority management, and divide the authority of team members into three types: editable, uploadable, and viewable, so as to facilitate the team to manage documents flexibly and meticulously. BRIEF DESCRIPTION OF THE DRAWINGS

[0065] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For ordinary technicians in this field, other drawings can also be obtained based on these drawings.

[0066] Figure 1 It is a system block diagram of a document management system based on data analysis according to an embodiment of the present invention;

[0067] Figure 2 It is a workflow diagram of a document management system based on data analysis according to an embodiment of the present invention;

[0068] Figure 3 This is another workflow diagram of a document management system based on data analysis according to an embodiment of the present invention. DETAILED DESCRIPTION

[0069] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0070] In addition, the described features, structures or characteristics may be combined in one or more example embodiments in any suitable manner. In the following description, many specific details are provided to provide a full understanding of the example embodiments of the present disclosure. However, those skilled in the art will appreciate that the technical solutions of the present disclosure may be practiced while omitting one or more of the specific details, or other methods, components, steps, etc. may be adopted. In other cases, well-known structures, methods, implementations or operations are not shown or described in detail to avoid obscuring various aspects of the present disclosure.

[0071] This embodiment provides a document management system based on data analysis. Figure 1 is a system block diagram of a document management system based on data analysis according to an embodiment of the present invention. Figure 1 As shown, the system includes a document information acquisition module, a document title automatic segmentation module, a document title part-of-speech tagging module, a document named entity recognition module and a document classification module;

[0072] The document information acquisition module is used to acquire document information, wherein the document information includes the document title;

[0073] The automatic segmentation module of the document title is used to predict the segmentation annotation sequence of the document title: convert the sentence corresponding to the document title into X=(x 1 , x 2 ,...,x n ), x i Represents the index value of the i-th character, i = 1, 2...n, input X into the character query table, and query each x i Convert it into a low-dimensional real number vector of fixed length, combine the low-dimensional real number vector into the input sentence LX(X), and input LX(X) into the bidirectional LSTM network to output the segmented annotation sequence;

[0074] The document title part-of-speech tagging module is used to predict the part-of-speech tagging sequence for the document title: the sentence corresponding to the document title is converted into an input vector using the FastText word embedding technology, and the input vector is input into the bidirectional LSTM network to output the part-of-speech tagging sequence;

[0075] The document named entity recognition module is used to annotate document titles with named entities;

[0076] The document classification module is used to classify documents based on document titles that contain a subscript tag sequence and a part-of-speech tag sequence and are annotated with named entities.

[0077] In summary, the present invention obtains document information, wherein the document information includes the document title; predicts the sub-writing tag sequence for the document title; predicts the part-of-speech tag sequence for the document title: converts the sentence corresponding to the document title into an input vector through the FastText word embedding technology, and inputs the input vector into a bidirectional LSTM network, outputs the part-of-speech tag sequence; performs named entity tagging on the document title; and classifies the document based on the document title containing the sub-writing tag sequence and the part-of-speech tag sequence and annotated with the named entity. The present invention can automatically classify, tag and retrieve the document based on the title of the document, improve the intelligent processing function of the document management system, achieve accurate classification, and improve the efficiency of document management.

[0078] It is worth noting that this embodiment provides a document management system based on data analysis and also includes: an online collaboration module for team members to upload data analysis attachments anytime and anywhere, and multiple people can collaborate in real-time editing, which reduces version conflicts and speeds up the data analysis process.

[0079] The social sharing module is used to share documents with other people. Others can view the document classification, document information, and document data analysis materials through the document link.

[0080] The permission hierarchical management module is used to divide the permissions of team members into three types: editable, uploadable, and viewable, making it convenient for the team to manage documents flexibly and meticulously.

[0081] In some embodiments, the document information may also include author name, journal name, publication year and other related document information, and the documents may be classified and managed according to the above process, which will not be elaborated here.

[0082] In some embodiments, by querying i The conversion into a low-dimensional real vector of fixed length specifically includes the following processes:

[0083] Through FastText word embedding technology, each x is transformed into i Convert to a low-dimensional real vector of fixed length:

[0084] The character sequence that constitutes the word is split into a set of n-grams, and the original word is added to form the basic input unit of model training. In the fastText model, the sum of the vectors of the subsequences in the n-gram set represents a word. For each x i Converted into a low-dimensional real vector u of fixed length t :

[0085]

[0086] Among them, Z j Represents xi The n-gram vector is obtained by table lookup of the word vector matrix, j=1,2...m, m is the number of n-gram vectors obtained by table lookup of the word vector matrix.

[0087] In some embodiments, LX(X) is input into a bidirectional LSTM network, and outputting a sub-written tag sequence specifically includes the following process:

[0088] The CRF layer of the LSTM network predicts the segmented annotation sequence:

[0089] S char (i) = f(W out h i +b out );

[0090]

[0091] Among them, S char (i) represents x i The probability distribution of the split annotation obtained by the bidirectional LSTM network, W out , b out are the mapping matrix and bias vector of the fully connected layer, f is the softmax function, A zy The transition matrix of the split-write annotation state, S(X, Y 1 , θ) represents a candidate tag sequence Y corresponding to the input character sequence X when considering the tag state transition probability and the split tag probability predicted by the bidirectional LSTM 1 The score of , where θ represents the model parameters;

[0092] Take the score S(X, Y) from all candidate annotation paths 1 ,θ) is the largest path as the segmentation annotation sequence.

[0093] In some embodiments, inputting the input vector into a bidirectional LSTM network and outputting a part-of-speech tag sequence includes the following process:

[0094] Predict the part-of-speech tag sequence through the CRF layer of the LSTM network:

[0095] S sylla (i) = f(W out h i +b out );

[0096]

[0097] Among them, S sylla (i) represents the probability distribution of part-of-speech tags obtained by the input vector through the bidirectional LSTM network, W out , bout are the mapping matrix and bias vector of the fully connected layer, f is the softmax function, A zx represents the transfer matrix of part-of-speech tagging, S(X, Y 2 ,δ) represents a candidate tag sequence Y corresponding to the input vector when considering the tag state transition probability and the probability of writing tags predicted by the bidirectional LSTM 2 The score of , where δ represents the model parameter;

[0098] Take the score S(X, Y) from all candidate annotation paths 2 ,δ) is the largest path as the part-of-speech tag sequence.

[0099] In some embodiments, performing named entity tagging on a document title specifically includes the following process:

[0100] Given a document title input sentence w = (w 1 , w 2 ,...,w n ), w i represents the i-th morpheme in the input sentence w of the document title, in, Represents the morphogen w i The gth syllable of ;

[0101] The document title is input into sentence w and processed by FastText word embedding technology to obtain the syllables represented by low-dimensional real number vectors The vector representing the g-th syllable in the i-th morpheme;

[0102] Will Input into the bidirectional LSTM network to get the last state of the forward LSTM network And the last state of the backward LSTM network

[0103]

[0104] and Represent the forward state and backward state operations respectively;

[0105] Will Concatenate into a vector γ i :

[0106] Among them, * represents the concatenation operation;

[0107] Will γ i Input into the fully connected network and output the named entity label of the morpheme through a softmax function.

[0108] In some embodiments, γ i The input is fed into a fully connected network, and the named entity label of the morpheme is output through a softmax function. The specific process includes the following:

[0109] pv(t)=softmax(W x γ i +b x );

[0110] p v (t) represents the probability distribution of named entity labels predicted using the syllable sequence in the morpheme, W x 、b x represents trainable parameters;

[0111] The objective function is:

[0112] Among them, t lk represents the kth named entity label of the lth morpheme. If the named entity of the lth morpheme belongs to the kth named entity category, then t lk is 1, otherwise, t lk is 0, represents the probability that the lth morpheme belongs to the kth named entity category, N is the number of morphemes, and C is the number of named entity categories.

[0113] In some embodiments, Figure 2 is a workflow diagram of a document management system based on data analysis according to an embodiment of the present invention, such as Figure 2 As shown, the document classification based on the document title containing the subscript tag sequence and the part-of-speech tag sequence and annotated with the named entity specifically includes the following steps:

[0114] Step 1: parse the annotation information:

[0115] Separate notations distinguish different clauses or phrases in a title;

[0116] Part-of-speech tagging provides the grammatical role each word plays in a sentence;

[0117] Named entity annotation points out important entities in the text;

[0118] Step 2: Extract key information:

[0119] Use named entity annotation to directly identify key entities in the title;

[0120] Combined with part-of-speech tagging, the verb phrases and noun phrases in the title are identified;

[0121] Step 3: Build a classification system:

[0122] Construct a classification system based on the research field, research type, and application object to which the literature belongs;

[0123] Step 4: Document classification:

[0124] Match the parsed and extracted key information with the classification system to determine the category to which the document title belongs.

[0125] Furthermore, the system also supports online collaboration and social sharing functions, supports uploading data analysis attachments, real-time collaborative editing by multiple people, reduces version conflicts, and shares documents with others through social software, so that others can view the document information.

[0126] In some embodiments, Figure 3 is another workflow diagram of a document management system based on data analysis according to an embodiment of the present invention, such as Figure 3 As shown in the figure, matching the parsed and extracted key information with the classification system to determine the category to which the document title belongs specifically includes the following steps:

[0127] Step S301: keyword matching, matching the extracted keywords with various category labels in the classification system; wherein, in the matching process, matching is performed based on the contextual information of the keywords in the title and their relationship with other words in the title;

[0128] Step S302: weight allocation. During the matching process, different key information is assigned different weights.

[0129] Step S303: Determine the category, and determine the category to which the document title belongs based on the matching result between the key information and the classification system.

[0130] The above embodiments can be implemented in whole or in part by software, hardware, firmware or any other combination. When implemented by software, the above embodiments can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, the process or function described in the embodiment of the present application is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions can be transmitted from one website site, computer, server or data center to another website site, computer, server or data center by wired (e.g., infrared, wireless, microwave, etc.). The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that contains one or more available media sets. The available medium can be a magnetic medium (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD), or a semiconductor medium. The semiconductor medium can be a solid-state hard disk.

[0131] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0132] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0133] In the several embodiments provided in the present application, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the units is only some logical function divisions. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0134] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0135] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A document management system based on data analysis, characterized in that: The system includes a document information acquisition module, a document title automatic segmentation module, a document title part-of-speech tagging module, a document named entity recognition module and a document classification module; The document information acquisition module is used to acquire document information, wherein the document information includes the document title; The automatic segmentation module of the document title is used to predict the segmentation annotation sequence of the document title: convert the sentence corresponding to the document title into X = (x1, x2, ..., x n ), x i Represents the index value of the i-th character, i = 1, 2...n, input X into the character query table, and query each x i Convert it into a low-dimensional real number vector of fixed length, combine the low-dimensional real number vector into the input sentence LX(X), and input LX(X) into the bidirectional LSTM network to output the segmented annotation sequence; The document title part-of-speech tagging module is used to predict the part-of-speech tagging sequence for the document title: the sentence corresponding to the document title is converted into an input vector using the FastText word embedding technology, and the input vector is input into the bidirectional LSTM network to output the part-of-speech tagging sequence; The document named entity recognition module is used to annotate document titles with named entities; The document classification module is used to classify documents based on document titles that contain a subscript tag sequence and a part-of-speech tag sequence and are annotated with named entities.

2. A document management system based on data analysis according to claim 1, characterized in that: Input X into the character lookup table, and query each x i The conversion into a low-dimensional real vector of fixed length specifically includes the following processes: Through FastText word embedding technology, each x is transformed into i Convert to a low-dimensional real vector of fixed length: The character sequence that constitutes the word is split into a set of n-grams, and the original word is added to form the basic input unit of model training. In the fastText model, the sum of the vectors of the subsequences in the n-gram set represents a word. For each x i Converted into a low-dimensional real vector u of fixed length t : Among them, Z j Represents x i The n-gram vector is obtained by table lookup of the word vector matrix, j=1,2...m, m is the number of n-gram vectors obtained by table lookup of the word vector matrix.

3. A document management system based on data analysis according to claim 1, characterized in that: Input LX(X) into the bidirectional LSTM network, and output the segmented label sequence, which specifically includes the following process: The CRF layer of the LSTM network predicts the segmented annotation sequence: S char (i)=f(W out h i +b out ); Among them, S char (i) represents x i The probability distribution of the split annotation obtained by the bidirectional LSTM network, W out 、b out are the mapping matrix and bias vector of the fully connected layer, f is the softmax function, A zy represents the transition matrix of the separate annotation state, S(X, Y1, θ) represents the score of a candidate annotation sequence Y1 corresponding to the input character sequence X when considering the annotation state transition probability and the separate annotation probability predicted by the bidirectional LSTM, where θ represents the model parameter; From all candidate annotation paths, the path with the largest score S(X, Y1, θ) is selected as the segmented annotation sequence.

4. A document management system based on data analysis according to claim 1, characterized in that: Inputting the input vector into the bidirectional LSTM network and outputting the optimal part-of-speech tag sequence includes the following process: Predict the part-of-speech tag sequence through the CRF layer of the LSTM network: S sylla (i)=f(W out h i +b out ); Among them, S sylla (i) represents the probability distribution of part-of-speech tags obtained by the input vector through the bidirectional LSTM network, W out , b out are the mapping matrix and bias vector of the fully connected layer, f is the softmax function, A zx represents the transfer matrix of the separate part-of-speech tagging, S(X, Y2, δ) represents the score of a candidate tag sequence Y2 corresponding to the input vector when considering the tag state transition probability and the separate tag probability predicted by the bidirectional LSTM, where δ represents the model parameter; From all candidate tagging paths, the path with the largest score S(X, Y2, δ) is taken as the part-of-speech tagging sequence.

5. A document management system based on data analysis according to claim 1, characterized in that: The named entity annotation of the document title specifically includes the following processes: Given a document title input sentence w = (w1, w2, ..., w n ), w i represents the i-th morpheme in the input sentence w of the document title, in, Represents the morphogen w i The gth syllable of ; The document title is input into sentence w and processed by FastText word embedding technology to obtain the syllables represented by low-dimensional real number vectors The vector representing the g-th syllable in the i-th morpheme; Will Input into the bidirectional LSTM network to get the last state of the forward LSTM network And the last state of the backward LSTM network and Represent the forward state and backward state operations respectively; Will Concatenate into a vector γ i : Among them, * represents the concatenation operation; Will γ i Input into the fully connected network and output the named entity label of the morpheme through a softmax function.

6. A document management system based on data analysis according to claim 5, characterized in that: Will γ i The input is fed into a fully connected network, and the named entity label of the morpheme is output through a softmax function. The specific process includes the following: p v (t)=softmax(W x γ i +b x ); p v (t) represents the probability distribution of named entity labels predicted using the syllable sequence in the morpheme, W x 、b x represents trainable parameters; The objective function is: Among them, t lk represents the kth named entity label of the lth morpheme. If the named entity of the lth morpheme belongs to the kth named entity category, then t lk is 1, otherwise, t lk is 0, represents the probability that the lth morpheme belongs to the kth named entity category, N is the number of morphemes, and C is the number of named entity categories.

7. A document management system based on data analysis according to claim 1, characterized in that: The document classification based on the document title containing the subscript tag sequence and the part-of-speech tag sequence and annotated with named entities specifically includes the following process: Step 1: parse the annotation information: Separate notations distinguish different clauses or phrases in a title; Part-of-speech tagging provides the grammatical role each word plays in a sentence; Named entity annotation points out important entities in the text; Step 2: Extract key information: Use named entity annotation to directly identify key entities in the title; Combined with part-of-speech tagging, the verb phrases and noun phrases in the title are identified; Step 3: Build a classification system: Construct a classification system based on the research field, research type, and application object to which the literature belongs; Step 4: Document classification: Match the parsed and extracted key information with the classification system to determine the category to which the document title belongs.

8. A document management system based on data analysis according to claim 1, characterized in that: Matching the parsed and extracted key information with the classification system to determine the category to which the document title belongs specifically includes the following steps: Keyword matching: Match the extracted keywords with the various category labels in the classification system; the matching process is based on the contextual information of the keywords in the title and their relationship with other words in the title; Weight allocation: During the matching process, different key information is assigned different weights; Determine the category: Determine the category to which the document title belongs based on the matching results of key information and the classification system.

Images