Project management chart automatic generation method and device, equipment and storage medium

Abstract

The invention provides a project management chart automatic generation method and device, equipment and a storage medium. The method comprises the steps that project management description information is acquired; generating code data corresponding to the description information based on a pre-trained generation model; and converting the code data into a project management chart of project management. According to the method, the description information of project management is input into the generation model, the generation model generates the corresponding code data according to the preset prompt template, and then the system converts the obtained code data into the corresponding project management chart. The corresponding project management chart can be automatically generated only by inputting the description information of project management, and the intelligent level of generating the project management chart is effectively improved.

Description

Technical Field

[0001] The present disclosure relates to the field of data processing technology, and in particular to a method, device, equipment and storage medium for automatically generating a project management chart. Background Art

[0002] Various charts are often used in project management to intuitively display various progress and data in project management. However, the generation of traditional project management charts requires manual input of a large amount of project management data to generate project management charts. In project management, the demand for automation and intelligence is increasing. How to automatically generate detailed project management charts by only inputting relevant descriptions of project management information has become a technical problem that technicians in this field need to solve urgently. Summary of the invention

[0003] In view of this, the present disclosure proposes a method, apparatus, device and storage medium for automatically generating a project management chart, which can automatically generate a detailed project management chart by only inputting relevant descriptions of project management information.

[0004] According to a first aspect of the present disclosure, a method for automatically generating a project management chart is provided, comprising: Get the description information of project management; Generate code data corresponding to the description information based on a pre-trained generation model; The code data is converted into a project management chart for the project management.

[0005] In a possible implementation manner, the project management description information is in any form of text, table, or voice.

[0006] In a possible implementation, the prompt template includes at least one of: information extraction requirements, information extraction precautions, generation model output format requirements, reference cases, and other format prompts.

[0007] In a possible implementation, before generating the corresponding code of the description information, the operation of training the generation model is also included; Wherein, when training the generation model, it includes: Get sample data; The generation model generates code data corresponding to the sample data according to a preset prompt template; The generation model is fine-tuned based on the code data to obtain a trained generation model.

[0008] In a possible implementation, when converting the code data into a project management chart corresponding to the description information, the code data is converted into the corresponding project management chart through a Mermaid interpreter.

[0009] According to a second aspect of the present disclosure, there is provided a device for automatically generating a project management chart, comprising: Data acquisition module, used to obtain descriptive information of project management; A code generation module, used to generate code data corresponding to the description information based on a pre-trained generation model; A chart conversion module is used to convert the code data into a project management chart for the project management.

[0010] In a possible implementation, the code generation module includes: A data acquisition module, used to acquire sample data; A model training module, used for the generation model to generate code data corresponding to the sample data according to a preset prompt template; The generation model determination module is used to fine-tune the generation model based on the code data to obtain a trained generation model.

[0011] According to a third aspect of the present disclosure, there is provided a device for automatically generating project management charts, comprising: a processor; and a memory for storing processor-executable instructions; wherein the processor is configured to execute the method described in the first aspect of the present disclosure.

[0012] According to a fourth aspect of the present disclosure, a non-volatile computer-readable storage medium is provided, on which computer program instructions are stored, wherein the computer program instructions, when executed by a processor, implement the method described in the first aspect of the present disclosure.

[0013] The present disclosure provides a method, device, equipment and storage medium for automatically generating a project management chart, the method comprising: obtaining project management description information; generating code data corresponding to the description information based on a pre-trained generation model; and converting the code data into a project management chart for project management. The above method inputs the description information of project management into the generation model, and the generation model generates the corresponding code data according to a preset prompt template, and then the system converts the obtained code data into the corresponding project management chart, so that the corresponding project management chart can be automatically generated by only inputting the description information of project management, which effectively improves the intelligent level of generating project management charts.

[0014] Further features and aspects of the present disclosure will become apparent from the following detailed description of exemplary embodiments with reference to the attached drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] The accompanying drawings, which are incorporated in and constitute a part of the specification, illustrate exemplary embodiments, features, and aspects of the disclosure and, together with the description, serve to explain the principles of the disclosure.

[0016] Figure 1 A flowchart showing a method for automatically generating a project management chart according to an embodiment of the present disclosure is shown; Figure 2 A schematic block diagram showing an apparatus for automatically generating a project management chart according to an embodiment of the present disclosure; Figure 3 A schematic block diagram of a device for automatically generating a project management chart according to an embodiment of the present disclosure is shown. DETAILED DESCRIPTION

[0017] Various exemplary embodiments, features and aspects of the present disclosure will be described in detail below with reference to the accompanying drawings. The same reference numerals in the accompanying drawings represent elements with the same or similar functions. Although various aspects of the embodiments are shown in the accompanying drawings, the drawings are not necessarily drawn to scale unless otherwise specified.

[0018] The word “exemplary” is used exclusively herein to mean “serving as an example, example, or illustration.” Any embodiment described herein as “exemplary” is not necessarily to be construed as preferred or advantageous over other embodiments.

[0019] In addition, in order to better illustrate the present disclosure, numerous specific details are given in the following specific embodiments. It should be understood by those skilled in the art that the present disclosure can also be implemented without certain specific details. In some examples, methods, means, components and circuits well known to those skilled in the art are not described in detail in order to highlight the subject matter of the present disclosure.

[0020] <Method Example> Figure 1 A flowchart of a method for automatically generating a project management chart according to an embodiment of the present disclosure is shown, which specifically includes: step S1100, obtaining description information of project management; step S1200, generating code data corresponding to the description information based on a pre-trained generation model; step S1300, converting the code data into a project management chart of project management.

[0021] Therefore, the present invention provides a method for automatically generating project management charts. By using a preset prompt template for the input project management description information, the generation model first extracts the required key information from the description information, then generates corresponding code data according to the prompt template, and then converts the code data into the corresponding project management chart to achieve automatic generation of the corresponding project management chart by only inputting the project management related description information, thereby effectively improving the intelligence level of generating project management charts. When generating a project management chart, the description information of the project management is first obtained, and then the description information is input into the generation model to generate corresponding code data.

[0022] Furthermore, when the project management description information is received, the project management description information is preprocessed according to the input form. The project management description information may be in any form of text, table or voice, and is not specifically limited.

[0023] Furthermore, before preprocessing the project management description information, the system will first determine the input form of the description information, and then use the corresponding preprocessing method according to the determined input form of the description information. For example, when the system determines that the project management description information is in text form, the text form preprocessing method is used; when the system determines that the project management description information is in table form, the table form preprocessing method is used; when the system determines that the project management description information is in voice form, the voice form preprocessing method is used.

[0024] When the description information is in text form, the preprocessing of the description information mainly includes at least one of error coding preprocessing and meaningless symbol preprocessing of the input text description information.

[0025] Specifically, when performing error coding preprocessing, an improved character coding detection tool can be used to automatically detect the coding format of the description information and identify possible error coding. Among them, the coding format includes a standard coding format and a mixed coding format. Error coding includes garbled characters and unrecognizable characters. Here, it should be pointed out that the standard coding format refers to a common single-mode coding, such as UTF-8 coding, GBK coding and ISO-8859 coding. A mixed coding format refers to a coding that combines two or more coding methods.

[0026] Therefore, when the description information in text form is preprocessed for error coding, the coding format of the coding information in text form is first determined. When it is determined that the coding format of the coding information is a standard coding format, the error coding identification method under the standard format is directly used to identify the error coding. When it is determined that the coding format of the coding information is a mixed coding format, the error coding identification method under the mixed format is directly used to identify the error coding.

[0027] Among them, the coding format of the coded information in text form is judged. When judging whether the coding format of the coded information in text form is a standard coding, it can be achieved by matching the coding method of the description information with the coding features of each coding method in the constructed standard coding list. Specifically, firstly, a standard coding list is constructed, and the standard coding list contains a variety of different standard coding methods and their corresponding coding features. Among them, the coding features of each coding method refer to the unique coding characteristics that are different from other codings.

[0028] Based on the constructed standard encoding list, the encoding in text form is traversed and matched with the encoding features of each standard encoding in the list. When the encoding method in text form does not completely match the encoding features in the list, it is judged that the encoding in text form is a mixed encoding format; when the encoding method in text form completely matches the encoding features in the list, it is judged that the encoding in text form is a standard encoding format. At this time, based on the corresponding relationship between the encoding features in the standard encoding list and each encoding, the specific encoding method of the description information in text form can be determined according to the matching encoding features.

[0029] For example, when judging whether the encoding format of the text-based encoding information is a standard encoding format, the following methods can be used: 1. Preliminary testing: Use an encoding detection tool (such as chardet) to scan the text and try to identify the encoding type. Compare the detection results with the standard encoding list (such as UTF-8, GBK, ISO-8859 series, etc.). If the detected encoding is in the list, it is marked as "standard encoding"; otherwise, it is marked as "non-standard encoding".

[0030] 2. Character-by-character verification: Decodes the text content character by character using the detected encoding. If illegal bytes or unparseable characters are found (such as values ​​outside the range of the character set), they are marked as non-standard encodings.

[0031] 3. Marking inspection: Detects the BOM (Byte Order Mark) in the text. The BOM of UTF-8 is EF BBBF. The BOM of UTF-16 is FE FF or FF FE.

[0032] If the text is missing the required BOM or has a mismatched BOM, it is flagged as a non-standard encoding.

[0033] 4. Contextual Validation: Perform contextual analysis on the text content to detect inconsistent characters (such as the presence of both ASCII and non-ASCII characters, which may be a sign of mixed encodings). Check whether the semantics are as expected. For example, the language content should conform to the declared encoding language range.

[0034] It should be noted that when determining whether the encoding format of the text-based encoded information is a standard format, any one of the above four determination methods may be used, or a combination of the four methods may be used in the above order, without specific limitation.

[0035] After obtaining the encoding method of the description information (i.e., the encoding format of the description information), when performing error encoding identification on the description information of the standard encoding format in combination with the encoding method, the error encoding is detected by a single encoding rule, wherein the encoding rule includes a legal byte mode and a character set encoding range.

[0036] Specifically, according to the determined encoding format, its legal byte pattern can be obtained, and the encoding of the description information is compared with its corresponding legal byte model in turn. The encoding that does not conform to the byte pattern of the current encoding format is the first error encoding. After the description information is identified as an error encoding by means of a byte pattern, the corresponding character set encoding range can be obtained based on the determined encoding format. According to the determined character set encoding range, the error encoding in other encodings other than the first encoding error in the description information is further identified. Specifically, the encodings other than the first encoding error in the description information are traversed to determine whether the other encodings are within the determined character set encoding range. When the traversed current encoding character is within the determined range, the current encoding character is the correct encoding; when the traversed current encoding character is not within the determined range, the current encoding character is the error encoding (i.e., the second error encoding). Finally, the first error encoding and the second error encoding are combined to form the final error encoding identified under the standard encoding format. Among them, the legal byte pattern and character set range corresponding to each encoding format under the standard encoding method are common knowledge in the art, and will not be elaborated on here.

[0037] For example, when identifying incorrect encoding of standard format encoded information, it includes: 1. Character-by-character parsing: Parse each character according to the declared encoding format to detect whether there is an illegal byte sequence. For example, the legal byte pattern of UTF-8 is: single byte: 0xxxxxxx, double byte: 110xxxxx 10xxxxxx, triple byte: 1110xxxx 10xxxxxx10xxxxxx. If it does not meet the above pattern, it is marked as an incorrect encoding.

[0038] 2. Illegal character detection: Checks if there are values ​​in the text that are outside the range of the character set. For example, the legal range of UTF-8 is Unicode U+0000 to U+10FFFF, and characters outside this range are marked as illegal.

[0039] 3. Frequency analysis: Perform frequency analysis on common characters in the text (such as English letters and punctuation marks). If the frequency distribution is abnormal (such as frequent occurrence of garbled characters), mark it as an incorrect encoding.

[0040] When it is determined that the encoding format of the description information is not a standard encoding format, it indicates that the description information is a mixed encoding format. Then, based on the fact that the encoding format is a mixed encoding format, the description information is subjected to mixed encoding error coding identification. When performing error coding identification, mixed coding requires segmented text parsing, and the encoding format is detected for each segment separately. Therefore, the identification under the mixed encoding format is more complicated. Specifically, the mixed coded description information is segmented, and then the encoding method of each segment of the description information is compared with each encoding feature in the standard encoding list to detect the encoding method of each segment of the description information. Then, each segment of the description information is traversed, and the error coding identification method of the above-mentioned standard coding is used to identify the error coding of each segment of the description information according to the corresponding confirmed encoding method. The traversal is completed, that is, the identification of the error coding under the mixed coding is completed. Among them, when segmenting the mixed coded description information, it can be segmented according to the text logical paragraphs or it can be segmented according to the fixed paragraph size, and there is no specific limitation.

[0041] For example, mixed coding identifies incorrect coding, including: 1. Segment processing: Divide the text into logical paragraphs or fixed-size segments. Perform encoding detection on each segment and record the possible encoding type of each segment.

[0042] 2. Conflict detection: Compare the encoding types detected in each segment. If the encodings of multiple segments are inconsistent (for example, some are UTF-8 and some are GBK), mark them as mixed encodings.

[0043] 3. Illegal character detection: Parse each piece of content character by character according to the detected encoding. Flag characters that cannot be decoded or values ​​that are out of range of the character set.

[0044] 4. Unified processing: For the detected mixed encoding parts, try to convert them into a single standard encoding (such as UTF-8).

[0045] It should be noted that, since the mixed coding error code recognition method can simultaneously perform coding format judgment and error code recognition, when preprocessing error codes in text form, the mixed coding error code recognition method can also be used to directly recognize error codes in text form.

[0046] Furthermore, after the error code in the description information is identified through the above steps, the error code can be corrected. When the error code is corrected, the error code is corrected based on the coding format of the description information confirmed in the process of identifying the error code.

[0047] Specifically, when the encoding format of the description information is a standard format and is the encoding format of UTF-8, the identified erroneous encoding is replaced or deleted. At this time, the erroneous encoding under the UTF-8 encoding format includes non-UTF-8 characters and invalid characters. It should be noted that when replacing or deleting the erroneous encoding of the description information under the UTF-8 encoding format, a method combining character mapping and context inference is used for replacement. Specifically, a character mapping table between common encoding errors and standard UTF-8 characters under the UTF-8 encoding format is first constructed. The erroneous encoding is replaced according to the character mapping table. Among them, when constructing a mapping table of standard UTF-8 characters corresponding to common encoding errors, the corresponding standard UTF-8 characters are confirmed according to the corresponding text represented by the encoding error under the common UTF-8 encoding format. Then the possible correct encoding of the current character is inferred based on the context and replaced. The erroneous encoding that cannot be replaced by the above methods is deleted.

[0048] For example, when replacing the identified error code, it includes: 1. Character Map: A character map is predefined to map common non-UTF-8 characters to standard Unicode characters. For example, 0x92 (a non-UTF-8 quotation mark) is mapped to the Unicode standard quotation mark '.

[0049] 2. Contextual Inference: Analyze the text around non-UTF-8 characters to infer their semantics. For example, if the character 0x91 appears between English words, it is probably an open quote.

[0050] 3. Character frequency analysis: Count the frequency of characters in similar texts and select the most common replacement characters.

[0051] It should be noted that when replacing the identified erroneous code, any one of the above three methods may be used, or the above methods may be combined to replace the erroneous code, without specific limitation.

[0052] When the encoding format of the description information is a non-UTF-8 encoding format (ie, an encoding format other than UTF-8 in the standard format, or a mixed encoding format), correction of the identified erroneous encoding may be achieved by using an advanced matching algorithm.

[0053] Specifically, when using an advanced matching algorithm to correct the error coding, it includes segmenting the description information according to the text logic, then traversing each text segment, using multiple encoding models to simultaneously parse the current text being traversed to generate the corresponding text, and based on the generated texts, selecting the model with the most coherent generated text and the lowest error rate as the encoding model for correcting the error coding of the current text, and using the text generated by the model as the corrected text of the current text. After the correction is completed, traverse the next text segment. When the traversal is completed, the correction of the error coding is completed.

[0054] For example, when using advanced matching algorithms to correct incorrect coding, including; 1. Segment analysis: Detect and record encoding features segment by segment according to the text logic segment.

[0055] 2. Coding model matching: Use multiple encoding models (such as UTF-8, GBK) for parallel parsing. Select the model that produces the most coherent text with the lowest error rate.

[0056] 3. Statistical correction: For the wrong characters that appear repeatedly in the text, their positions and contexts are counted, and the most frequently used replacement characters are used for correction.

[0057] Here, it should be noted that when correcting the error code by the advanced matching algorithm, it is possible that the correction cannot be successfully performed. Therefore, after correcting the error code by the advanced matching algorithm to obtain the correction result, a step of judging whether the correction is successful can also be included according to the correction result.

[0058] After judging that the correction is successful, it indicates that the correction result is the correct encoding, and the correct encoding obtained by correction can be directly replaced with the previous incorrect encoding. When judging that the correction is unsuccessful, it indicates that the correction result cannot be directly used to replace the previous incorrect encoding. At this time, the encoding format of the description information is converted to a unified preset format (i.e., UTF-8 encoding format), and then the description information converted to the preset format is identified and corrected for the incorrect encoding.

[0059] Among them, it should be pointed out that when judging the correction result, it is achieved through text verification, target generation verification and semantic consistency verification. Specifically, text verification refers to decoding the corrected description information character by character according to a determined encoding format. When the description information can be decoded completely normally, it means that the error coding has been corrected successfully; when the description information cannot be decoded completely normally, it means that there are still error codings that have not been corrected. At this time, the description information is converted into a unified preset format to identify and correct the error coding. Target generation verification refers to generating the target chart or code with the corrected text to check whether the generated result meets expectations. Semantic consistency verification refers to using the NLP model to verify whether the corrected text is semantically coherent.

[0060] It should be noted that when judging the correction result, any one of the above methods may be used, or the above methods may be combined to judge the correction result, without specific limitation.

[0061] It should be noted that when identifying incorrect codes for description information converted into UTF-8 encoding format, the method for identifying incorrect codes and the method for handling the identified incorrect codes are consistent with those of the UTF-8 encoding method under the above-mentioned standard encoding format, and will not be elaborated on here.

[0062] At the same time, according to the foregoing, the preprocessing of the descriptive information in text form also includes the preprocessing of meaningless symbols. Specifically, when preprocessing the meaningless symbols, it is necessary to first determine the meaningless symbols. That is, by identifying all the symbols in the descriptive information, the meaningless symbols are then determined from all the identified symbols. Here, it needs to be explained that the meaningless symbols refer to symbols that have no practical meaning or function in the text. Among them, the identification method of all the symbols of the descriptive information is common knowledge in the art and will not be elaborated on here.

[0063] After all symbols in the description information are identified, meaningless symbols are further identified. When a meaningless symbol is identified, the meaningless symbol is identified through a context-aware intelligent symbol identification mechanism. The context-aware intelligent symbol identification mechanism refers to analyzing the context of the symbol, analyzing the text structure before and after the symbol, and whether the symbol contributes to the information expression in the context to determine whether the symbol is a meaningless symbol.

[0064] It should be noted that when performing context analysis on a symbol, the text structure before and after the symbol is first analyzed. Specifically, the syntactic dependency relationship of the words before and after the symbol is extracted, and a syntax tree of the current symbol is constructed according to the extracted syntactic dependency relationship as the first syntax tree, and the currently constructed syntax tree structure is recorded. Then the current symbol is deleted, and a syntax tree is constructed for the sentence after the current symbol is deleted as the second syntax tree, and the corresponding syntax structure is recorded. The syntax structures of the first syntax tree and the second syntax tree are compared. When the syntax structures of the first syntax tree and the second syntax tree are consistent, it means that the current symbol has no meaning for the text structure. When the syntax structures of the first syntax tree and the second syntax tree are inconsistent, it means that the current symbol has meaning for the text structure.

[0065] For example, the text structure analysis of symbols includes: 1. Dependency extraction: Use dependency analysis tools (such as SpaCy) to extract the syntactic dependencies between the words before and after the token.

[0066] 2. Syntax tree parsing: Build a syntax tree and check whether the symbol changes the syntactic structure of the sentence.

[0067] 3. Continuity test: Regenerate the syntax tree after deleting the symbol and compare the syntactic changes.

[0068] After judging that the current symbol has no meaning for the text structure, further judge whether the current symbol contributes to the information expression in the context to confirm the meaningless symbol. When judging that the symbol contributes to the information expression in the context, the current symbol is not a meaningless symbol; when judging that the symbol does not contribute to the information expression in the context, the current symbol is a meaningless symbol. Among them, when judging whether the symbol contributes to the information expression in the context based on the text structure before and after the analyzed symbol, by deleting the current symbol, judge whether the current symbol has an impact on the information expression. Specifically, record the information expression when the current symbol exists, then delete the current symbol, and record the information expression after deleting the current symbol. Then compare the information expression before and after the deletion of the current symbol. Judge whether the information expression before and after the deletion of the current symbol is consistent. When the information expression before and after the deletion of the current symbol is consistent, judge that the current symbol does not contribute to the information expression, and at this time judge that the current symbol is a meaningless symbol; when the information expression before and after the deletion of the current symbol is inconsistent, judge that the current symbol contributes to the information expression, and at this time judge that the current symbol is a meaningful symbol.

[0069] For example, when judging whether the current symbol is a meaningless symbol, it includes: 1. Semantic segmentation test: Reparse the text after removing the symbol and test whether it causes semantic incoherence.

[0070] 2. Separation function check: Checks that symbols clearly separate contexts (such as separating items in a manifest).

[0071] 3. Verification of syntactic integrity: Check the effect of symbols on syntactic structure (e.g., whether omitting separators causes ambiguity).

[0072] Furthermore, when deleting confirmed meaningless symbols, the deletion rate of meaningless symbols can also be adjusted according to the text type of the description information. Specifically, when it is determined that the text type of the description information is a table type text, the deletion rate of tab characters is reduced; when it is determined that the text type of the description information is a free text type, the deletion rate of redundant spaces and line breaks is increased. Wherein, when determining the text type of the description information, it is determined based on the text structure of the description information, specifically, whether the text structure of the description information has a fixed row and column structure. When the text structure of the description information is a fixed row and column structure, the text type of the description information is determined to be a table type; when the text structure of the description information is not a fixed row and column structure, the text type of the description information is determined to be a free text type. Wherein, when detecting the text structure of the description information, a feature extraction algorithm can be used for detection.

[0073] Furthermore, after confirming the meaningless symbols, the meaningless symbols can be deleted by calculating the weights of the meaningless symbols, and then comparing them with the weight threshold to determine the meaningless symbols to be deleted. Among them, when calculating the weights of the meaningless symbols, first evaluate and score the three aspects of the position of the current meaningless symbol in the sentence, the contribution to the sentence expression, and the contribution to the sentence structure to obtain the corresponding three scores. Then the three scores are calculated according to the contribution ratio to obtain the weight value of the current meaningless symbol. Based on the weights of the meaningless symbols obtained, they are compared with the corresponding weight thresholds. Those below the weight threshold are the final meaningless symbols, which are deleted; otherwise, the current meaningless symbols are retained.

[0074] For example, the position, semantics, and format contributions of meaningless symbols are scored as follows: Location Contribution Score: Weighting: The importance of symbols at the beginning, end, and middle of a sentence decreases in turn. For example, a symbol at the beginning of a sentence may serve as a paragraph separator and has a higher score.

[0075] Scoring rules: When the symbol appears at the beginning of a sentence, the score is 1.0; when the symbol appears at the end of a sentence, the score is 0.8; when the symbol appears in the middle of a sentence, the score is 0.5.

[0076] Example: Text: "Project Phase: Design Phase, Start Time." The ":" symbol at the beginning of the sentence has a score of 1.0.

[0077] Semantic contribution scoring: Weighting basis: Contribution of detected symbols to the semantic coherence of the context.

[0078] Scoring rules: If the semantics becomes incoherent after deleting the symbol, the score is 1.0; if the semantics changes slightly after deleting the symbol, the score is 0.7; if the semantics does not change after deleting the symbol, the score is 0.3.

[0079] Example: Text: "The design phase, followed by the development phase." After deleting "," the semantics changes slightly, and the score is 0.7.

[0080] Format contribution score: Weighting is based on whether the symbol contributes to text formatting (such as table alignment, paragraph separation).

[0081] Scoring rules: When the symbol affects table alignment, the score is 1.0; when the symbol affects paragraph separation, the score is 0.8; when the symbol has no formatting contribution, the score is 0.2.

[0082] Based on the three dimension scores obtained above, the corresponding contribution ratios are determined according to the dimension importance, for example, the position contribution ratio is determined to be 30%, the semantic contribution ratio is 50%, and the format contribution ratio is 20%. The weight of the current symbol is calculated according to the determined contribution ratio and the corresponding dimension score. It can be expressed as: in, Indicates the score of the current meaningless symbol position, Indicates the score of the semantic contribution of the current meaningless symbol, Indicates the score of the current meaningless symbol format contribution.

[0083] Based on the obtained weights of each meaningless symbol, it is compared with the corresponding weight threshold. The symbol below the weight threshold is the final meaningless symbol and is deleted; otherwise, the current meaningless symbol is retained. When determining the weight threshold, the global importance distribution of various symbols in the description information is analyzed through large-scale corpus statistics. Based on the global importance distribution of various symbols, the global importance weight values ​​of various symbols are calculated, and then the average value is calculated, which is the weight threshold of the meaningless symbols in the current description information.

[0084] Through the above-mentioned recognition and processing of meaningless symbols, we can continuously learn the recognition and processing methods of meaningless symbols in the historical texts that have been processed. When processing similar texts in the future, we can directly identify and delete meaningless symbols by learning that the symbols are meaningless in specific contexts.

[0085] When the description information is in a table format, the content in the table is converted into complete text information and then processed in text format. In one possible implementation, the empty cells in the table are preprocessed before the text content of the table is identified. Specifically, all cells in the table are traversed and identified to determine whether the traversed cells are empty. When the traversed and identified cells have text content recorded in them, it is determined that the current traversed cell is not empty, and the traversal continues; when the current traversed cell does not have text content recorded in it, it is determined that the current traversed cell is empty, and the currently traversed empty cell is specially processed. Among them, special processing includes inserting a default value or marking it as empty, and there is no specific limitation.

[0086] After empty cell preprocessing, the preprocessed table is structured and parsed to identify the header cells and data cells in the table. The header cells record the project phase, start date, and end date. The data cells record detailed information on project management.

[0087] It should be noted that when performing structured parsing of a table, a keyword library for the title cell is first constructed, which contains keywords for the project phase, start date, and end date. Then, based on the constructed keyword library for the title cell, it is determined whether the cell in the table is a cell that records the title. Specifically, each cell in the table is traversed, and the text recorded in the cell is matched with the keywords in the keyword library. If the text content recorded in the traversed cell matches the keywords in the keyword library, the current cell is determined to be a title cell; otherwise, if the text content recorded in the traversed cell does not match the keywords in the keyword library, the current cell is determined to be a data cell.

[0088] Then, based on the obtained title cells, data cells and the contents recorded in each cell, each cell is subjected to correlation analysis to obtain the correlation between each project phase and time, and each project phase recorded in the data cell is matched with time one by one according to the obtained correlation to form descriptive information in text form. Among them, when each cell is subjected to correlation analysis to obtain the correlation between each project phase and time, it includes firstly traversing the data cells according to the fixed row and column structure and title cells of the table based on the identified title cells, data cells and the fixed row and column structure of the table, extracting multiple data cells corresponding to each title, that is, extracting each data cell corresponding to the project phase recorded by the project phase title cell, the data cell corresponding to the start date recorded by the start date title cell, and the data cell corresponding to the end date recorded by the end date title cell. Then, according to the fixed row and column structure, the data cells recording each project phase under the project phase title cell are matched one by one with each start date and end date recorded under the time title cell, and finally the correlation between each project phase and time is obtained.

[0089] Based on the correlation analysis of each unit, the correlation relationship between each project stage and time is obtained, and each project stage in the table is associated with the corresponding time, and finally a textual description is formed. Finally, the textual description is preprocessed and input into the generation model.

[0090] When the description information is in speech form, speech recognition is performed on the input speech to identify it as text. The recognized description information in text form is then pre-processed in text form before being input into the generation model. When the description information in speech form is processed into text, any speech-to-text conversion method can be used, without specific limitation. The present disclosure preferably uses an ASR speech-to-text model.

[0091] After obtaining the description information of the project management, the generation model will generate the corresponding code data according to the preset prompt template. The generation model can use any natural language processing model such as BERT, GPT-3 or a large language model based on prompt fine-tuning technology, without specific limitation, and preferably uses a large language model based on prompt fine-tuning technology.

[0092] Before generating the corresponding code data of the description information, a preset prompt template is required, wherein the preset prompt template includes: at least one of information extraction requirements, information extraction precautions, generation model output format requirements, reference cases and other format prompts.

[0093] Specifically, information extraction requirements are some requirements and prompts for the generation model to extract key information based on the input project management description information, and the setting of the requirements and prompts must be specific and clear to ensure that the generation model can accurately understand and extract the required information. For example, when generating a project management Gantt chart, for the provided project management description information, the prompt template will be set to: Please extract the various project phases, time nodes and milestones from the following project information: Project phases: design, development, testing, deployment Time nodes: start date, end date Milestone: A significant event or deliverable Information extraction precautions are to avoid fabrication and hallucination when the generative model extracts and outputs information. Information extraction precautions will be set in the prompt template. For example, the precautions can be set as: Note: Please extract strictly according to the information provided, and do not add or fabricate any information that is not provided. If a piece of information is missing, please mark it as "not provided".

[0094] The model output format requirement is to enable the generated model to output the input description information according to the required format requirements. For example, when using the generated model to generate project management Gantt chart code data, the output format requirement in the prompt template can be set as: Please output the Markdown code of the Gantt chart in the following format: ```mermaid gantt title Project Gantt Chart dateFormat YYYY-MM-DD section Project Phase Design:des1, 2023-01-01, 30d Development: dev1, after des1, 60d Test:test1, after dev1, 30d Deployment: dep1, after test1, 15d Reference cases are used to ensure that the generated model understands the description information correctly and the output format is accurate. The prompt template sets the corresponding reference cases. For example, when using the generated model to generate the corresponding code data for the project management Gantt chart, the prompt template can set the reference case as: Sample input: Project design begins on January 1, 2023 and lasts 30 days. The development phase follows the design phase and lasts 60 days. The testing phase begins after development is completed and lasts 30 days. The deployment phase begins after testing is completed and lasts 15 days.

[0095] Sample output: ```mermaid gantt title Project Gantt Chart dateFormat YYYY-MM-DD section Project Phase Design:des1, 2023-01-01, 30d Development: dev1, after des1, 60d Test:test1, after dev1, 30d Deployment: dep1, after test1, 15d Other formatting hints refer to the relevant hints carried in the input information during the process of generating model recognition input information. In order to avoid the generation model from confusing the hint information with the project information, corresponding formatting requirements are made. For example, the information input into the generation model will embed the description information into the hint template and input it into the generation model together. At this time, the input information will contain hint information and project information. In order to distinguish between hint information and project information, the hint information will be set to start with hint, and the project information will be set to start with project information. Therefore, when setting the hint template, the formatting hint will be set as follows: the hint information will start with "hint:", and the project information will start with "information:". Please do not confuse these two parts.

[0096] After the prompt template is constructed based on the above processing, the corresponding prompt template will be constructed according to the generated project management chart type, and the constructed prompt template and the generated chart type will be stored one by one for subsequent model generation calls. For generating different types of project management charts, the method of constructing the prompt template is consistent with the above construction method, and no further elaboration will be given here.

[0097] After obtaining the preset prompt template, firstly search for the corresponding prompt template according to the type requirement of the generated project management chart carried in the description information. Before searching for the corresponding prompt template, firstly determine the type of the generated project management chart.

[0098] In one possible implementation, when determining the type of project management chart to be generated, first define a chart keyword pool, a chart type pool, and the correspondence between chart types and keywords. The correspondence between chart types and keywords can be flexibly specified and is not specifically limited. The chart keyword pool contains keywords that mark the generated chart type, and the chart type pool contains chart types for generating the chart. Then, a keyword library corresponding to the chart type is constructed based on the correspondence between the chart type and the keyword.

[0099] Since there are many types of charts, this application takes the process of constructing a corresponding keyword library for one chart type as an example to illustrate the process of constructing a corresponding keyword library for each chart type. Specifically, the chart type is taken out from the chart type pool, and then the keywords of the corresponding chart type are taken out from the keyword pool according to the correspondence between the keywords and the chart type to form a keyword library corresponding to the current chart type. It should be noted that the method for constructing a corresponding keyword library for each chart type is consistent with the above method, and no further elaboration is given here.

[0100] Based on the keyword library corresponding to each type of chart obtained, it is determined whether the description information contains keywords of the type of the logo generated chart. Specifically, all the words in the description information are traversed, and the current word traversed is matched with the keywords stored in the keyword pool. When there is a keyword matching the current word traversed, it is determined that the description information contains keywords of the type of the logo generated icon; when there is no keyword matching the current word traversed, it is determined that the description information does not contain keywords of the type of the logo generated icon.

[0101] When determining that the description information contains a keyword indicating the type of icon to be generated, the chart type of the generated chart is determined based on the stored correspondence between the chart type and the keyword.

[0102] When it is determined that the description information does not contain a keyword indicating the type of icon to be generated, the description information is contextually analyzed using natural language processing technology to infer the type of icon to be generated for project management. Among them, using natural language processing technology to contextually analyze the description information to infer the type of chart to be generated for project management is common knowledge in the art and will not be elaborated on here.

[0103] It should be noted that when determining that the description information does not contain a keyword indicating the type of icon to be generated, contextual analysis is performed on the description information to determine the type of chart to be generated, and the type of chart to be generated may also be inferred based on historical behavior and preference.

[0104] After determining the chart type for generating project management, search for the corresponding prompt template according to the determined chart type. At this time, search for the corresponding prompt template based on the confirmed chart type, and embed the description information into the prompt template as a complete prompt input into the generation model. The generation model will extract the key information corresponding to the generated project management chart from the description information according to the complete prompt. Among them, when the complete prompt is input into the generation model, the prompt information starts with prompt and the project information starts with project information. At this time, the project information and prompt information are first identified, and then the identified project information is extracted according to the prompt information.

[0105] In a possible implementation, when extracting key information, first define a phase pool and a milestone pool corresponding to the chart type. The phase pool contains each project phase related to the generated chart type, and the milestone pool contains each project milestone related to the generated chart type. The project phases in the phase pool and the milestones in the milestone pool can be flexibly determined, and there is no specific limitation.

[0106] Based on the defined phase pool corresponding to each generated chart type, the corresponding project phase keywords are extracted from the description information. Specifically, the corresponding defined phase pool is searched according to the determined chart type, and then all words in the description information are matched with the keywords in the phase pool, that is, all words in the description information that match the words in the keywords are screened, and the first keyword that is the same as the keyword in the phase pool is screened. For example, if the keyword is the start phase, and the description information is described as the start of the project plan, then start is used as the first keyword.

[0107] Then, based on the keywords of each project stage in the stage pool, it is determined whether the first keyword is a standard keyword. Specifically, each word in the first keyword is completely matched with the keywords in the stage pool, and the words that can completely match the keywords in the stage pool are determined to be standard keywords. At this time, the project stage keywords are directly extracted. When each word in the first keyword cannot completely match the keywords in the stage pool, it is determined that the current first keyword is not a standard keyword. At this time, context analysis is used to infer the standard project stage.

[0108] After extracting the key information of the project phase, it is necessary to extract the key information of the milestone. The extraction method used is consistent with the extraction process of the key information of the project phase mentioned above, and will not be elaborated here.

[0109] After extracting the project phases and milestones, time nodes are extracted. Before extracting the time nodes, it is necessary to determine whether the time nodes are time nodes in a standard format. The time in a standard format refers to the time format required in the prompt information. Specifically, when determining whether the time is in a standard format, first identify all descriptions of the time in the description information, and then determine whether the description of the identified time is in a standard format based on the time format required in the prompt information. When the currently identified time is a time that meets the time format requirements, it is determined that the current time is in a standard format, and the current time information is directly extracted; when the currently identified time does not meet the time format requirements, it is determined that the current time is not in a standard format. At this time, a natural language time parsing tool is used to convert the non-standard time into a standard time format, and then the time that meets the time format requirements is extracted.

[0110] After extracting the key information of project phases, milestones and time of project management, context analysis is performed on the description information to obtain the correspondence between each project phase, milestone and time, and the extracted project phases, milestones and time are matched one by one according to the obtained correspondence.

[0111] For example, the description is entered as: "The project design begins on January 1, 2023 and lasts for 30 days. The development phase follows and lasts for 60 days." The extracted time nodes are: Design phase: start date is 2023-01-01, end date is 2023-01-30 Development phase: start date is 2023-02-01, end date is 2023-03-30 Based on the key information extracted by the above generation model, the generation model will generate corresponding code data that meets the requirements according to the prompt template. The code data is Markdown code.

[0112] Furthermore, after finding the corresponding template according to the generated project management chart type, the prompt template can be adjusted according to the input project management project type as the final prompt template. The description information is then embedded into the prompt template as a complete prompt and input into the generation model to generate corresponding code data. The description information contains keywords about the project type.

[0113] Specifically, when adjusting the built prompt template for different project types, firstly, based on the predefined project phase pool and milestone pool, the required phase and milestone requirement diagrams are built according to the project type, and the corresponding typical phases are taken from the phase pool and the corresponding milestones are taken from the milestone pool according to the built requirement diagram, and they are saved according to the project type. Among them, when building the requirement diagram corresponding to the project type, it can be flexibly determined without limitation.

[0114] For example, software type projects: The typical stages are identified as: requirements analysis, design, development, testing, and release.

[0115] The milestones are determined as: requirement freeze, development completion, test pass, and product release.

[0116] Building Type Project: The typical stages are identified as: design, planning, construction, and acceptance.

[0117] The milestones are: design approval, foundation completion, main structure completion, and acceptance.

[0118] When adjusting the prompt template, the prompt template is adjusted according to the item type carried by the description information, including adjusting the information extraction requirements and the output format.

[0119] Among them, the adjustment of information extraction requirements includes: stage description adjustment and milestone adjustment. Specifically, when adjusting the stage description, find the corresponding saved typical stage according to the project type, and replace the stage description prompt in the prompt template. For example, for software type projects, the template contains prompts of "design stage" and "test stage"; if the input project type is a construction type project, find the corresponding saved project stage according to the construction project type, and adjust the "design stage" and "test stage" in the prompt template to "construction stage" and "acceptance stage". When adjusting milestones, find the corresponding saved milestones according to the project type, and replace the milestone description prompt in the prompt template. For example, when the prompt template is a scientific research project type, the milestone description prompt is "experimental design completed". If the input project type is a construction project, find the saved milestones according to the construction project type, and replace the "experimental design completed" in the prompt template with "foundation completed" in the construction project.

[0120] Output format adjustment includes: output structure adjustment and time node refinement. Specifically, when adjusting the output structure, the output format requirements in the prompt template are adjusted by identifying and analyzing the parallelism of tasks in different project stages of different project types. For example, if the system recognizes that "development" and "testing" are carried out simultaneously in a software project, the output format requirements in the prompt template are adjusted to output "development" and "testing" in parallel.

[0121] When making detailed adjustments to the time nodes, by identifying the time information in the description information, the relationship between the identified time nodes is determined to determine whether the time is carried out according to a strict schedule or has a flexible time period. If there is a strict schedule, the time nodes are output according to the strict schedule; if there is a flexible time period between the identified time nodes, the output time node requirements are adjusted accordingly so that the output time requirement is the time period in which the corresponding node occurs.

[0122] After adjusting the prompt template, you can get the prompt template corresponding to the project management chart generated for different project types. The obtained prompt template is stored, so that when the corresponding project management chart is generated for the same project type in the future, the adjusted prompt template can be directly called to generate the corresponding project management chart.

[0123] After obtaining the adjusted final prompt template, the description information is embedded into the prompt template and input into the generation model as a complete prompt. The generation model will generate corresponding code data based on the prompt template.

[0124] It should be noted that before the generative model generates code data corresponding to the description information according to the prompt template, the generative model must be trained first.

[0125] In one possible implementation, training the generative model includes: First, obtain sample data. Specifically, collect and organize the corresponding project management data for the type of chart to be generated. For example, for generating a project management Gantt chart, collect and organize the corresponding project management description information, including project phases, time nodes, milestones and other related information.

[0126] Second, the generation model extracts the key information required to generate the project management chart from the relevant description information according to the preset prompt template, and then the generation model generates the code data corresponding to each sample data that meets the requirements according to the prompt template.

[0127] Specifically, the generation model will extract a prompt based on the project information in the prompt template to perform information extraction operations, obtain at least one key information among the project phases, time nodes, and milestones, and then perform time parsing and formatting operations based on a prompt based on the time node formatting requirements in the prompt template, and convert the time nodes into start dates and end dates that comply with the Mermaid specification. Subsequently, a task dependency identification operation is performed based on a prompt based on the task parallelism requirements in the prompt template to determine the sequential relationship and parallel situation of the tasks. Finally, the generation model fills the extracted information into a Markdown code framework that complies with the Mermaid structure based on the output format prompt to generate the final code data. Among them, the output format requirement prompt includes task parallelism requirements and time node formatting requirements.

[0128] Third, a trained generative model is obtained based on the code data. Specifically, the generative model will generate code data corresponding to each sample data according to the preset prompt template. At this time, the model is fine-tuned according to the code data output by the generative model to make the code data output by the model more accurate. Finally, the generative model with the most accurate code data after fine-tuning is determined as the final generative model for generating the project management chart.

[0129] Among them, fine-tuning of the generated model can adopt any one of supervised learning, calculation of loss function and setting of hyperparameters, without specific limitation.

[0130] Furthermore, when using supervised learning to fine-tune the generative model, the project management description information and the corresponding code data are provided to the generative model. The generative model is continuously trained based on the provided description information and corresponding code data, and the parameters of the generative model are adjusted so that accurate description information corresponding to the code data is finally generated to obtain the final generative model.

[0131] When fine-tuning the generative model by calculating the loss function, the description information of each sample is passed through the generative model to generate the corresponding code data, which is then compared with the corresponding target code data to calculate the loss function of the generated code data and the target code data. The calculation loss function can be expressed as: Among them, L represents the loss value, N represents the total number of samples, C represents the number of output categories of each sample, i represents the i-th sample, and j represents the j-th category of the i-th sample. represents the true distribution of sample i (target code data), the value is 1 for the target category and 0 for the non-target category. represents the generation distribution of sample i (generated code data), and the value is the probability generated by the model. is the weight of the j-th category. Then, based on the obtained loss function, the back propagation algorithm is used to pass the loss value to the generation model. At this time, the generation model adjusts the generation parameters (i.e., adjusts the parameters of the generation model) through the gradient descent algorithm, so that the loss value gradually decreases, and finally the generation model with the most accurate generation code data is obtained. Among them, using back propagation and gradient descent to adjust the parameters of the model is common knowledge in the field, and will not be elaborated on here.

[0132] When fine-tuning the generative model by setting hyperparameters, the corresponding fine-tuning hyperparameters of the fine-tuning generative model are first set, including: setting the learning rate of the generative model and the number of fine-tuning rounds. Specifically, the learning rate determines the amplitude of each update of the model. Setting it too high will cause the model to be unstable, and setting it too low will slow down the training speed. It is preferably set to 1e-5 or 1e-6. The number of fine-tuning rounds is set according to the size of the data set. Too few rounds may lead to inaccurate generation results, and too many rounds may lead to overfitting. It is preferably set to 3-10 rounds. Next, the generative model is trained according to the set hyperparameters, and the parameters are continuously adjusted so that the final generative model generates accurate description information corresponding to the code data to obtain the final generative model.

[0133] Furthermore, when determining the final generated model, a scoring model or a method for calculating accuracy may be used, without specific limitation.

[0134] Specifically, when the scoring model is used to determine the final generation model, the fine-tuned generation model is used to generate code data, and then the code snippets in the code data are scored according to the scoring dimensions. Among them, the scoring method can be manual scoring or automatic scoring, and there is no specific limitation. The scoring dimensions include: grammatical correctness, information accuracy, and format standardization. It should be noted that grammatical correctness refers to whether the generated Markdown code conforms to the grammatical rules of Mermaid. Information accuracy refers to whether the project phase, time node, milestone and other information in the generated code snippet accurately reflects the input description information. Format standardization refers to whether the format of the generated code is beautiful and easy to understand.

[0135] After obtaining the generated code scores, the weighted average score is calculated, and the generation model with the highest weighted average score is selected as the final generation model. The weighted average score can be expressed as: In the formula, i represents the number of scoring items, represents the score of the i-th item (such as grammar score, format score, accuracy score, etc.), represents the weight of the i-th score, and n represents the total number of score dimensions.

[0136] Among them, a corresponding weight will be set for each score, and the weight can be set according to the importance of the scoring dimension. Preferably, the weight of important scoring dimensions is set to 0.4, and the weight of other scoring dimensions is set to 0.2.

[0137] When the final generation model is determined by the calculation accuracy method, the code data generated by the generation model is compared with the target code data to calculate the accuracy of the code data generated by the generation model. The calculation accuracy can be expressed by the formula: In the formula, the number of correctly generated elements includes at least one of the project phase, time node, and milestone. The total number of elements represents all key information in the generated target code.

[0138] The final generation model is determined by calculating the accuracy of the code data generated by the generation model. The higher the accuracy, the more accurate the code data generated by the generation model. The generation model with the highest accuracy is finally determined as the final generation model for generating the corresponding code data for the generated project management description information.

[0139] For example, in order to generate a corresponding generative model for a project management Gantt chart, a large amount of sample data will be collected first, that is, descriptive information of project management, including at least one of the project phases, time nodes, and milestones. Then, according to the key information required by the Gantt chart, the corresponding prompt template will be preset. The generative model will first extract the key information of each sample data according to the prompt template, and then generate the corresponding code data. The generative model will be fine-tuned according to the generated code data, and the generated code data will be made more accurate by continuously adjusting the model parameters. Finally, the generative model with the most accurate generated code data will be determined as the final generative model for generating the Gantt chart.

[0140] After obtaining the trained generation model, the trained generation model will generate the corresponding code data of the project management description information according to the preset prompt template. The process of the generation model generating the corresponding code is consistent with the process of the generation model generating the corresponding code data in the above-mentioned model training, and will not be described in detail here. After obtaining the code data corresponding to the project management description information, the code data is converted into a project management chart corresponding to the description information. Specifically, the Mermaid interpreter is called to convert the generated code data into the corresponding project management chart. Among them, the conversion process of converting the Markdown code into the corresponding chart through the Mermaid interpreter is common knowledge in the art and will not be described in detail here.

[0141] It should be noted that the system also has customization capabilities, which customizes the part that displays the project management chart after parsing the code data, including the visual modification and color configuration of the project management chart, so that the displayed project management chart is clearer and easier to understand and can highlight the key points.

[0142] Therefore, the present disclosure provides a method for automatically generating a project management chart, including: obtaining project management description information; generating code data corresponding to the description information based on a pre-trained generation model; and converting the code data into a project management chart of the project management. The above method inputs the description information of the project management into the generation model, and the generation model generates the corresponding code data according to the preset prompt template, and then the system converts the obtained code data into the corresponding project management chart, so that the corresponding project management chart can be automatically generated by only inputting the description information of the project management, which effectively improves the intelligent level of generating project management charts.

[0143] <Device Example> FIG. 1 is a schematic block diagram of a device for automatically generating a project management chart according to an embodiment of the present disclosure. Figure 2 As shown, the device 100 includes: a data acquisition module 110, a code generation module 120, and a chart conversion module 130. The data acquisition module 110 is used to acquire the description information of the project management; the code generation module 120 is used to generate the code data corresponding to the description information based on the pre-trained generation module; and the chart conversion module 130 is used to convert the code data into a project management chart of the project management.

[0144] <Equipment Embodiment> Figure 3 FIG. 1 is a schematic block diagram of a device for automatically generating a project management chart according to an embodiment of the present disclosure. Figure 3 As shown, the project management chart automatic generation device 200 includes: a processor 210 and a memory 220 for storing executable instructions of the processor 210. The processor 210 is configured to implement any of the above-mentioned project management chart automatic generation methods when executing the executable instructions.

[0145] Here, it should be noted that the number of processors 210 can be one or more. At the same time, in the project management chart automatic generation device 200 of the embodiment of the present disclosure, an input device 230 and an output device 240 may also be included. Among them, the processor 210, the memory 220, the input device 230 and the output device 240 may be connected through a bus or in other ways, which are not specifically limited here.

[0146] The memory 220 is a computer-readable storage medium that can be used to store software programs, computer executable programs, and various modules, such as the program or module corresponding to the project management chart automatic generation method of the embodiment of the present disclosure. The processor 210 executes various functional applications and data processing of the project management chart automatic generation device 200 by running the software program or module stored in the memory 220.

[0147] The input device 230 may be used to receive input numbers or signals. The signals may be key signals related to user settings and function control of the device / terminal / server. The output device 240 may include display devices such as display screens.

[0148] <Storage Medium Embodiment> According to a fourth aspect of the present disclosure, a non-volatile computer-readable storage medium is also provided, on which computer program instructions are stored. When the computer program instructions are executed by the processor 210, any of the above-mentioned methods for automatically generating project management charts is implemented.

[0149] The embodiments of the present disclosure have been described above, and the above description is exemplary, not exhaustive, and is not limited to the disclosed embodiments. Many modifications and changes will be apparent to those of ordinary skill in the art without departing from the scope and spirit of the described embodiments. The selection of terms used herein is intended to best explain the principles of the embodiments, practical applications, or technical improvements in the market, or to enable other persons of ordinary skill in the art to understand the embodiments disclosed herein.

Claims

1. A method for automatically generating a project management chart, characterized in that: include: Get the description information of project management; Generate code data corresponding to the description information based on a pre-trained generation model; The code data is converted into a project management chart for the project management.

2. The method according to claim 1, characterized in that The project management description information is in any form of text, table, or voice.

3. The method according to claim 1, characterized in that The generation model generates code data corresponding to the description information according to a preset prompt template.

4. The method according to claim 3, characterized in that The prompt template includes: at least one of: information extraction requirements, information extraction precautions, generation model output format requirements, reference cases and other format prompts.

5. The method according to claim 1, characterized in that Before generating the corresponding code of the description information, the operation of training the generation model is also included; Wherein, when training the generation model, it includes: Get sample data; The generation model generates code data corresponding to the sample data according to a preset prompt template; The generation model is fine-tuned based on the code data to obtain a trained generation model.

6. The method according to claim 1, characterized in that When converting the code data into a project management chart corresponding to the description information, the code data is converted into the corresponding project management chart through a Mermaid interpreter.

7. A device for automatically generating project management charts, characterized in that: include: Data acquisition module, used to obtain descriptive information of project management; A code generation module, used to generate code data corresponding to the description information based on a pre-trained generation model; A chart conversion module is used to convert the code data into a project management chart for the project management.

8. The device according to claim 7, characterized in that The code generation module includes: A data acquisition module, used to acquire sample data; A model training module, used for the generation model to generate code data corresponding to the sample data according to a preset prompt template; The generation model determination module is used to fine-tune the generation model based on the code data to obtain a trained generation model.

9. A project management chart automatic generation device, characterized in that: include: processor; a memory for storing processor-executable instructions; Wherein, the processor is configured to implement the method described in any one of claims 1 to 6 when executing the executable instructions.

10. A non-volatile computer-readable storage medium having computer program instructions stored thereon, characterized in that: When the computer program instructions are executed by a processor, the method according to any one of claims 1 to 6 is implemented.

Images