A blockchain-based incremental timeline engineering information extraction method and system

Abstract

The present invention belongs to the technical field related to the field of engineering safety management, and discloses a method and system for extracting engineering information based on incremental timeline based on blockchain, the method comprising: using a shard classification model to identify the engineering project to which the engineering text uploaded by the user belongs and sending the transaction task to the shard of the engineering project to which it belongs; the shards execute their respective transaction tasks, including: first using an event clustering model to identify the engineering theme to which the engineering text belongs in the corresponding engineering project, then using an event classification model to identify the engineering event to which the engineering text belongs in the corresponding engineering theme, inserting the engineering text into the timeline of the engineering event to which it belongs according to the recording time of the engineering event, integrating the timeline to generate a timeline summary of the engineering event to which it belongs: integrating all the timeline summaries in the shard to generate a shard summary of the corresponding engineering project; integrating all the shard summaries to generate a global summary. Through the above method, the required engineering information can be quickly extracted.

Description

Technical Field

[0001] The present invention belongs to the technical field related to the field of construction project safety management, and more specifically, to a blockchain-based incremental timeline engineering information extraction method and system. Background Art

[0002] In the field of construction engineering, engineering monitoring reports, construction logs, etc. are important reference documents in project management, which are crucial to the property safety of the project and the personal safety of the staff. The construction unit and the monitoring unit will obtain the progress of various events of the project based on these documents and give suggestions on measures to be taken. Such text information is very easy to be tampered with, and it will be difficult to trace the source and hold accountable once an engineering accident occurs. Blockchain technology, as a decentralized, tamper-proof, transparent and open distributed ledger technology, provides a strong guarantee for data security. Therefore, using blockchain technology to manage engineering texts can improve the authenticity and integrity of data and prevent data from being tampered with or forged.

[0003] However, the main challenge facing these project reference documents is the huge amount of information they contain. They describe many events, but not all of the information is needed by stakeholders such as government agencies and construction units. In addition, these documents are time-sensitive, and the events described in them will change as the documents are updated, which undoubtedly reduces the utilization and efficiency of relevant information and data.

[0004] Therefore, it is necessary to propose a method based on blockchain to quickly extract the required engineering information. Summary of the invention

[0005] In view of the above defects or improvement needs of the prior art, the present invention provides an incremental timeline engineering information extraction method and system based on blockchain, which aims to quickly extract engineering information based on blockchain.

[0006] To achieve the above object, the present invention provides an incremental timeline engineering information extraction method based on blockchain, wherein the blockchain comprises multiple shards, and different shards are used to process different engineering projects;

[0007] The method comprises:

[0008] Using the shard classification model to identify the engineering project to which the engineering text uploaded by the user belongs and sending the transaction task of the engineering text to the shard corresponding to the engineering project;

[0009] Different shards execute their own transaction tasks in parallel. The transaction tasks include generating a timeline summary and a shard summary. The process of generating the timeline summary includes: firstly using an event clustering model to identify the engineering theme to which the engineering text belongs in the corresponding engineering project, then using an event classification model to identify the engineering event to which the engineering text belongs in the corresponding engineering theme, inserting the engineering text into the timeline of the engineering event to which it belongs according to the recording time of the engineering event, integrating the timeline to generate a timeline summary of the engineering event to which it belongs, and the timeline is obtained by sorting different description files for the same engineering event according to time; the process of generating the shard summary includes: integrating all the timeline summaries in the shard to generate a shard summary of the corresponding engineering project;

[0010] Integrate all shard summaries to generate a global summary;

[0011] Among them, the fragmentation classification model, event clustering model and event classification model are all trained neural network models.

[0012] Optionally, the sharding classification model is obtained by training on the basis of the Chinese BERT model, followed by adding a fully connected layer and a softmax function in sequence; the event clustering model is obtained by training on the basis of the Chinese BERT model, followed by adding an MLP layer downstream of the model; the event classification model is obtained by training the Llama3-7B model.

[0013] Optionally, the process of using an event classification model to identify an engineering event to which the engineering text belongs within a corresponding engineering theme, inserting the engineering text into a timeline of the engineering event to which the engineering text belongs according to the recording time of the engineering event, and integrating the timeline to generate a timeline summary of the engineering event to which the engineering text belongs includes:

[0014] Fine-tuning is performed on the basis of the Llama3-7B model, so that after the Llama3-7B model identifies the engineering event to which it belongs, it inserts the engineering text into the timeline of the engineering event to which it belongs according to the recording time of the engineering event according to the first fine-tuning instruction, integrates the timeline according to the second fine-tuning instruction to generate a timeline summary of the engineering event to which it belongs, and integrates all timeline summaries in the segment according to the third fine-tuning instruction to generate a segment summary.

[0015] Optionally, before inserting the project text into the timeline, first determine whether the current project text is related to the project event in the current timeline and whether the project text has available information. If the current project text is related to the project event in the current timeline and has available information, the current project text is inserted into the timeline, otherwise, it is not inserted.

[0016] Optionally, the event line of each engineering event is stored in the storage space of the corresponding shard in the form of a list. When generating a summary, the list content of the corresponding engineering event is directly sorted to obtain a timeline summary of the corresponding engineering event.

[0017] Optionally, each shard contains at least members of various stakeholders of the project, and each member is assigned to a corresponding shard according to his role and relevance in the project.

[0018] Optionally, when a transaction within a shard is completed, the resulting transaction data is broadcast to other shards through cross-shard synchronization operations, so that all nodes participating in the corresponding project on all shards can access the latest transaction data.

[0019] Optionally, the trigger condition for updating the timeline summary is that the user submits the project text to the blockchain, the trigger condition for updating the shard summary is that the timeline summary is updated, and the global summary is updated according to a preset period.

[0020] Optionally, the engineering documents include monitoring reports, construction logs, material purchase lists, and rectification orders.

[0021] The present invention also provides an incremental timeline engineering information extraction system based on blockchain, characterized in that it includes a blockchain and a smart contract deployed on the blockchain, wherein the blockchain includes multiple shards, and different shards are used to process different engineering projects, and the smart contract includes a transaction shard contract, a transaction formation contract, and a transaction supplement contract, wherein:

[0022] The transaction sharding contract is used to use the sharding classification model to identify the engineering project to which the engineering text uploaded by the user belongs and send the transaction task of the engineering text to the shard corresponding to the engineering project;

[0023] The transaction formation contract is used to execute the transaction task when the shard receives the transaction task. Different shards execute their own transaction tasks in parallel. The transaction task includes generating a timeline summary and a shard summary. The process of generating the timeline summary includes: firstly using the event clustering model to identify the engineering theme to which the engineering text belongs in the corresponding engineering project, and then using the event classification model to identify the engineering event to which the engineering text belongs in the corresponding engineering theme, inserting the engineering text into the timeline of the engineering event to which it belongs according to the recording time of the engineering event, integrating the timeline to generate the timeline summary of the engineering event to which it belongs, and the timeline is obtained by sorting different description files for the same engineering event according to time; the process of generating the shard summary includes: integrating all the timeline summaries in the shard to generate the shard summary of the corresponding engineering project;

[0024] The transaction supplement contract is used to integrate all shard summaries to generate a global summary;

[0025] The fragmentation classification model, event clustering model and event classification model are all trained neural network models.

[0026] In general, compared with the prior art, the above technical solutions conceived by the present invention mainly have the following beneficial effects:

[0027] 1. The present invention adopts a blockchain transaction sharding method. Sharding can improve the speed of summary generation and processing and reduce network load. Since each shard only processes a part of the transaction, it can reduce the data transmission and calculation amount of the entire network and reduce the burden on the node. By distributing transactions to each shard for processing, each node can process more transactions, thereby improving the throughput of the entire network;

[0028] 2. The present invention uses blockchain for data management and stores summary information in blocks, making the summary content reliable and enabling information sharing and traceability. At the same time, due to the characteristics of blockchain, the timeline of the same event is preserved intact and will not be tampered with, which helps relevant personnel to manage and grasp the development and evolution of events from an overall perspective, making it easier to understand the relationship and impact between events. At the same time, the clustering operation of events classifies and organizes related events, making it easier for users to quickly find and manage related events. If a problem occurs, it is more convenient to trace the development process of related events;

[0029] 3. The present invention generates summaries at three levels: project events, project themes, and project globality. This synergy helps to obtain event information in more detail, avoid information confusion or duplication, help to form a clear timeline summary and structure, and improve the utilization and accuracy of text information;

[0030] 4. The present invention uses model automatic recognition to quickly cluster and classify events in monitoring daily reports, construction logs and other documents, construct an incremental timeline of the same event, and generate a summary suitable for evolving real-time text, thereby achieving timely updates on the progress of various project events, reducing manual workload and the time to obtain effective information. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] Figure 1 It is a schematic diagram of the structure of blockchain transaction sharding according to an embodiment of the present invention.

[0032] Figure 2 This is a flowchart of the steps of a blockchain-based incremental timeline engineering information extraction method in one embodiment of the present invention.

[0033] Figure 3 It is an information processing flow chart of incremental timeline engineering information extraction in one embodiment of the present invention. DETAILED DESCRIPTION

[0034] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention. In addition, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0035] The present invention is based on blockchain network for data management. Therefore, it is necessary to first create a blockchain network, define nodes according to the actual participation of the project, and form a network topology, such as Figure 1 The figure shows a schematic diagram of the structure of blockchain transaction sharding in one embodiment of the present invention. In the present invention, in order to improve the efficiency of transaction processing, the blockchain sharding technology is adopted to divide the blockchain into multiple shards, each shard is used to execute transactions of different engineering projects, and all shards can conduct transactions in parallel. For example, a blockchain network can be created based on hyperledger fabric, a single channel can be created and all nodes can be added to the channel, a chain code containing sharding logic can be deployed on a single channel, a classification rule based on transaction content can be implemented in the chain code, a data isolation layer can be established inside the chain code, and transaction sharding can be implemented by storing transactions in different data tables.

[0036] In one embodiment, each work slice contains at least one member of each relevant party of the project, and each member is assigned to a designated slice according to his role and relevance in the project. In this way, even if information transactions are conducted by slice, each slice has the participation of personnel from each relevant party, which can increase the supervision of slice transactions. For example, the slice corresponding to the foundation pit excavation project has construction personnel, supervision personnel, survey and design personnel, government regulators, construction party supervisors, etc.

[0037] Based on the above blockchain, the present invention proposes the following blockchain-based incremental timeline engineering information extraction method.

[0038] like Figure 2 FIG. 1 is a flowchart of the steps of the incremental timeline engineering information extraction method based on blockchain in one embodiment of the present invention. Figure 3 The figure shows an information processing flow chart of incremental timeline engineering information extraction in one embodiment of the present invention. The steps are described in detail below.

[0039] Step S1: Use the shard classification model to identify the engineering project to which the engineering text uploaded by the user belongs and send the transaction task of the engineering text to the shard corresponding to the engineering project.

[0040] Specifically, engineering documents can be monitoring reports, construction logs, material purchase lists, rectification orders and other documents. These documents have specific dates, and for the same event, the progress is updated frequently, and there are potential mutual impacts between these events. For example, "purchased materials have not been delivered" will affect the progress of the project. These are all information that relevant personnel need to pay attention to.

[0041] When a node user submits a project text to the blockchain, it triggers the blockchain to classify and identify the project text to identify the project to which the project text belongs. Transaction tasks of different project projects can be distributed to different shards for processing, so that transaction events can be traded and data shared by project management stakeholders within their shards. When multiple different transaction tasks arrive at the same time, different shards can be executed in parallel, which can improve transaction efficiency.

[0042] For example, if the project includes foundation pit excavation, structural engineering quality supervision and surrounding environment monitoring, the blockchain can be divided into three shards, with shard 1 responsible for the transaction tasks of foundation pit excavation, shard 2 responsible for the transaction tasks of structural engineering quality supervision, and shard 3 responsible for the transaction tasks of surrounding environment monitoring, and so on. For example, the text content "The 5th support between the 1st to 7th frame has been dismantled, the 1st to 7th frame bottom plate has been completed, the 4th support of the 1st frame has begun to be removed, the 4th support of the 5th frame has begun to be removed, and the 4th support between the 5th and 6th frames is being removed" and "The casting of the A3-4 bottom plate of the data center has been completed" both belong to the foundation pit excavation shard.

[0043] The shard classification model used in this step is a trained neural network model. In the early stage, it can be trained with engineering texts with engineering project tags so that it can automatically identify the engineering projects to which the engineering texts belong after training, thereby distributing the transaction tasks of the engineering texts to the corresponding shards.

[0044] In a specific embodiment, the shard classification model is obtained by sequentially adding a fully connected layer and a softmax function to the Chinese BERT model. Through training, the parameters of the fully connected layer and the softmax function are optimized so that the shard classification model can correctly classify the transaction tasks into the corresponding shards based on the input text. The Chinese BERT model can extract the content in the engineering text and identify the shard to which it belongs through the fully connected layer and the softmax function.

[0045] Step S2: Different shards execute their own transaction tasks in parallel. The transaction tasks include generating timeline summaries and shard summaries. The process of generating timeline summaries includes: firstly using the event clustering model to identify the engineering theme to which the engineering text belongs in the corresponding engineering project, and then using the event classification model to identify the engineering event to which the engineering text belongs in the corresponding engineering theme, inserting the engineering text into the timeline of the engineering event to which it belongs according to the recording time of the engineering event, integrating the timeline to generate the timeline summary of the engineering event to which it belongs, and the timeline is obtained by sorting different description files for the same engineering event according to time; the process of generating shard summaries includes: integrating all the timeline summaries in the shards to generate shard summaries of the corresponding engineering projects.

[0046] After the new data is stored in the database with tags, event transactions and shard transactions are performed in the shard to which it belongs, thereby generating a timeline summary and a shard summary.

[0047] Specifically, when a shard receives a transaction task, it first generates a timeline summary based on the engineering text and then regenerates it into a shard summary.

[0048] Specifically, the event clustering model is first used to identify the engineering topics to which the engineering text belongs in the corresponding engineering project. Each engineering project can have multiple engineering topics, such as Figure 3 Each slice has multiple clusters. For example, under the engineering project of structural engineering quality supervision, it is divided into pile monitoring topics, surface settlement monitoring topics, etc. For example, when the engineering text is identified as an engineering project of structural engineering quality supervision, the event clustering model is further used to identify which topic under structural engineering quality supervision the engineering text belongs to, such as: the text content "The positive change rate of the inclinometer monitoring points ATCX37 and ATCX38 in the 1st to 7th area is greater than 1mm in the excavation depth range. The wall inclinometer hole A-CX32 (28m) has accumulated 77.81 mm, which has exceeded the early warning control (70mm), and the change rate is temporarily stable." belongs to the pile monitoring topic.

[0049] The event clustering model used in this step is a trained neural network model. In the early stage, it can be trained using engineering texts with engineering topic tags so that after training, it can automatically identify the engineering topic to which the engineering text belongs, so as to cluster engineering events under the corresponding engineering topic.

[0050] In a specific embodiment, the event clustering model is based on the Chinese BERT model, and an MLP layer is added downstream of the model. The MLM self-supervised learning technology is used to optimize the parameters of the MLP layer through training, so that the event clustering model can output correct clustering results based on the input text.

[0051] After identifying the engineering theme, the event classification model is used to identify the engineering events to which the engineering text belongs in the corresponding engineering theme. Each engineering theme can have multiple engineering events. For example, "1st to 7th floor slab reinforcement binding completed" belongs to the engineering event "1st to 7th floor slab progress".

[0052] After the corresponding engineering event is identified, the engineering text is inserted into the timeline of the engineering event according to the recording time of the engineering event, and the timeline is integrated to generate a timeline summary of the engineering event. The timeline is obtained by arranging the engineering time according to the order of time.

[0053] For example, for the event "Progress of the 1st to 7th floorboards", its daily progress is sorted by time to form the following timeline of the event:

[0054] "On October 21, the reinforcement of the 1st to 7th bottom slabs was completed.";

[0055] "On October 23, the pouring of the 1st to 7th floor slabs was completed.";

[0056] By analogy, the texts belonging to the event "Progress of the 1st to 7th baseboards" are sorted by time to obtain the timeline of the event.

[0057] After the timeline of the project event is updated, the timeline is integrated to generate a timeline summary of the project event to obtain information about the project event.

[0058] After the timeline summary is updated, all the timeline summaries in the shard are integrated to generate a shard summary of the corresponding project to obtain information about the project topic.

[0059] The characteristic of the timeline summary of the timeline is that the summary will reflect the progress status of a specific event, and it can be updated in time when changes occur; the characteristic of the summary of the slice is that the summary reflects the progress status of a certain project theme, which involves the timelines of events within the slice affecting each other. For example, the summary of a slice describes that "the foundation pit excavation work will be completed in 10 days", if this slice receives a new event "the original planned cement quantity is insufficient, it has been used up, and it needs to be purchased again, and the construction period is postponed by one day", then the summary of this slice will be updated to "due to insufficient cement, the construction period is postponed, and the foundation pit excavation work will be completed in 11 days".

[0060] The event classification model used in this step is also a trained neural network model. In the early stage, it can be trained using engineering texts with event tags so that it can automatically identify the events to which the engineering texts belong after training, so as to insert the contents of the engineering texts into the timeline of the corresponding events.

[0061] In a specific embodiment, the event classification model adopts the Llama3-7B model, and the parameters of the Llama3-7B model are optimized through training so that the event classification model can identify the engineering event to which the input text belongs.

[0062] Furthermore, after the engineering event is identified, the subsequent operations of updating the timeline, updating the timeline summary, and updating the fragment summary can be implemented by fine-tuning the Llama3-7B model. Specifically, the Llama3-7B model can be fine-tuned multiple times in the form of a template. The first fine-tuning instruction is used to insert the description content in the text into the timeline of the event according to the event after identifying the specific engineering event, and obtain an updated timeline. The second fine-tuning instruction is used to integrate the content of the timeline after obtaining the updated timeline, and obtain the timeline summary of the corresponding event. The timeline summary reflects the development process of the event. The third fine-tuning instruction is used to integrate all the timeline summaries in the fragment to obtain an updated fragment summary. The fine-tuned Llama3-7B model can be called an event timeline classification model. The function of the event timeline classification model is to insert the input text content into the timeline of the corresponding event and update the timeline summary and fragment summary of the event.

[0063] Furthermore, before inserting the project text into the timeline, it is first determined whether the current project text is related to the project event in the current timeline and whether the project text has available information. If the current project text is related to the project event in the current timeline and has available information, the current project text is inserted into the timeline, otherwise, it is not inserted. The determination can be completed by the Llama3-7B model.

[0064] In one embodiment, the timeline of each event is stored in the storage space of the corresponding shard in the form of a list. When generating a summary, the list content of the corresponding event is directly sorted to obtain the timeline summary of the corresponding event. Specifically, couchdb can be used as a database for storing data.

[0065] In a specific embodiment, the first fine-tuning instruction can be expressed as:

[0066] You will get a list of events, in chronological order, that refer to an event in the following timeline:

[0067] {timeline_events}

[0068] Consider the following new event:

[0069] {new_event}

[0070] If this new event follows the given list of events in the same timeline (i.e., the new event is about the same event), respond “yes”;

[0071] If the new event is not related to an event in the given timeline, reply “No, not relevant”;

[0072] If the new event is a duplicate or redundant (i.e., it repeats information from a previous event in the given timeline), respond “No, it is a duplicate”;

[0073] If a new event is not informative (i.e., it is generic or expresses some ideas but does not add new information to the events in the given timeline), then respond “No, it is not informative”;

[0074] # Answer.

[0075] In a specific embodiment, the second fine-tuning instruction can be expressed as:

[0076] You will be given a start event defined under ##Seed, you may have incoming information related to the start event under ##Timeline, write a summary combining the start event and the incoming information, if no incoming information is provided, summarize the start event;

[0077] ##Seed

[0078] {first tweet in timeline}

[0079] ##Timeline

[0080] {rest of the tweets in the timeline}

[0081] ##Summary.

[0082] In a specific embodiment, the third fine-tuning instruction can be expressed as:

[0083] You will get the timeline summary defined under ##Change with the new event added, you may get other timeline summaries within the same shard under ##TimelineSummary, write a shard summary combining all the above timeline summaries, if no incoming information is provided, summarize the timeline summary to which the new event belongs;

[0084] ##Change

[0085] {Timeline summary with new events added}

[0086] ##TimelineSummary

[0087] {Summary of other timelines in the shard}

[0088] ##Summary.

[0089] Step S3: Integrate all the fragment summaries to generate a global summary.

[0090] In this step, based on the updated shard summaries of each shard, a global summary, that is, information of the entire project, can be obtained by integrating the shard summaries.

[0091] Specifically, considering the large amount of computation required for global information integration, its update cycle can be set reasonably. For example, once the user provides text, it can trigger the update of the timeline summary, which in turn triggers the update of the shard summary. The global summary can be updated once a day. For example, after identifying the completion of the last transaction in the shard every day, the transaction partition data is locked, data sharing is completed through cross-shard transactions, and the data is organized into a global summary, which is then submitted and recorded on the blockchain.

[0092] Specifically, cross-shard transactions include: cross-shard data synchronization, that is, a cross-shard data synchronization module is included between different shards. When a transaction occurs within a shard, a cross-shard synchronization operation is automatically triggered to transfer transaction data from one shard to another. The synchronization module ensures data consistency across shards through a distributed consensus algorithm, ensuring that all nodes participating in the shard can access the latest transaction data to avoid data conflicts.

[0093] In one embodiment, the Llama3-7B model may also be fine-tuned by applying a fourth fine-tuning instruction to arrange all the shard summaries to generate an updated global summary.

[0094] In a specific embodiment, the fourth fine-tuning instruction can be expressed as:

[0095] You will get the shard summaries with updates defined under ##Alteration, you may get all other shard summaries under ##ShardSummary, write a global summary combining all the above shard summaries, if no incoming information is provided, aggregate the updated shard summaries;

[0096] ##Alteration

[0097] {Updated shard summary}

[0098] ##ShardSummary

[0099] {Summary of all other shards}

[0100] ##Summary.

[0101] The present invention also provides an incremental timeline engineering information extraction system based on blockchain, which includes a blockchain and a smart contract deployed on the blockchain. The blockchain includes multiple shards, and different shards are used to process different engineering projects. The smart contract includes a transaction shard contract, a transaction formation contract, and a transaction supplement contract.

[0102] The transaction sharding contract is used to use the sharding classification model to identify the engineering project to which the engineering text uploaded by the user belongs and send the transaction task of the engineering text to the shard corresponding to the engineering project.

[0103] Specifically, the content in the engineering text is identified and classified through the shard classification model, and the shard o1 with the highest similarity is obtained. The content is marked with the o1 shard and stored in couchdb, and the shard o1 is triggered to execute the transaction task.

[0104] The transaction formation contract is used to execute the transaction task when the shard receives the transaction task. Different shards execute their own transaction tasks in parallel. The transaction task includes generating a timeline summary and a shard summary. The process of generating a timeline summary includes: first using the event clustering model to identify the engineering theme to which the engineering text belongs in the corresponding engineering project, and then using the event classification model to identify the engineering event to which the engineering text belongs in the corresponding engineering theme, inserting the engineering text into the timeline of the engineering event according to the recording time of the engineering event, integrating the timeline to generate a timeline summary of the engineering event, and the timeline is obtained by sorting different description files for the same engineering event according to time; the process of generating a shard summary includes: integrating all the timeline summaries in the shard to generate a shard summary of the corresponding engineering project.

[0105] Specifically, shard o1 queries and obtains content with o1 shard tag in couchdb and identifies the engineering theme to which it belongs in the corresponding engineering project through the event clustering model, and then identifies the engineering events in the corresponding engineering theme through the event classification model, and then retrieves similar events and their timelines to determine the relationship between the current tweet and the retrieved timeline. After determining the timeline to which the event belongs, add the timeline header of the current event progress according to the timeline to which it belongs, and summarize the timeline to form a new timeline summary. Then, automatically trigger the generation of shard summary, that is, query the event transactions sorted by timeline in shard o1, extract the summary of each timeline, form a new summary as a whole, form a shard transaction, and submit it to the shard to which it belongs. In the specific execution, the operations of updating the timeline, updating the timeline summary, and updating the shard summary can all be implemented by fine-tuning on the basis of the Llama3-7B model, and specifically, the fine-tuning instructions can be applied to the Llama3-7B model multiple times in the form of a template.

[0106] The transaction supplement contract is used to integrate all shard summaries to generate a global summary.

[0107] Specifically, when the transaction supplement contract is triggered, the global shard summary is locked and integrated to form a global summary.

[0108] The fragmentation classification model, event clustering model and event classification model are all trained neural network models.

[0109] It can be understood that various contracts in the system can be used to implement corresponding transactions in the embodiment. Transactions are divided into events, shards, and globals. For example:

[0110] An event transaction is a summary of the event information and the transaction submission. That is, after a new event is correctly classified into its own timeline, the event timeline classification model is fine-tuned to generate a timeline summary for this timeline.

[0111] Shard transactions are summaries of the latest summaries of each timeline in the shard; that is, after a new event is correctly classified into its own timeline, the event timeline classification model is fine-tuned and the summaries of all timelines in the shard are input into the model to generate a summary of the shard;

[0112] The global transaction is a summary formed by summarizing all shards; that is, the shard summaries of all shards are input into the event timeline classification model after fine-tuning instructions to generate a global summary.

[0113] Event transactions and shard transactions are generated once each time a new event is submitted, while global transactions are generated once a day.

[0114] For details of the above transactions, please refer to the introduction of the incremental timeline engineering information extraction method, which will not be repeated here.

[0115] The technical features of the above embodiments can be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification. It should be noted that "in one embodiment", "for example", "for example", etc. of the present invention are intended to illustrate the present invention, rather than to limit the present invention.

[0116] The above embodiments only express several implementation methods of the present invention, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the patent application. It should be pointed out that, for those of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present invention, and these all belong to the protection scope of the present invention.

Claims

1. A blockchain-based incremental timeline engineering information extraction method, characterized in that: The blockchain includes multiple shards, and different shards are used to process different engineering projects; The method comprises: Using the shard classification model to identify the engineering project to which the engineering text uploaded by the user belongs and sending the transaction task of the engineering text to the shard corresponding to the engineering project; Different shards execute their own transaction tasks in parallel. The transaction tasks include generating a timeline summary and a shard summary. The process of generating the timeline summary includes: firstly using an event clustering model to identify the engineering theme to which the engineering text belongs in the corresponding engineering project, then using an event classification model to identify the engineering event to which the engineering text belongs in the corresponding engineering theme, inserting the engineering text into the timeline of the engineering event to which it belongs according to the recording time of the engineering event, integrating the timeline to generate a timeline summary of the engineering event to which it belongs, and the timeline is obtained by sorting different description files for the same engineering event according to time; the process of generating the shard summary includes: integrating all the timeline summaries in the shard to generate a shard summary of the corresponding engineering project; Integrate all shard summaries to generate a global summary; Among them, the fragmentation classification model, event clustering model and event classification model are all trained neural network models.

2. The incremental timeline engineering information extraction method based on blockchain as claimed in claim 1 is characterized in that: The segmentation classification model is obtained by training by adding a fully connected layer and a softmax function on the basis of the Chinese BERT model; The event clustering model is obtained by adding an MLP layer to the downstream of the Chinese BERT model for training; The event classification model is obtained by training the Llama3-7B model.

3. The incremental timeline engineering information extraction method based on blockchain as claimed in claim 2 is characterized in that: The method of using the event classification model to identify the engineering event to which the engineering text belongs in the corresponding engineering theme, inserting the engineering text into the timeline of the engineering event according to the recording time of the engineering event, and integrating the timeline to generate a timeline summary of the engineering event includes: Fine-tuning is performed on the basis of the Llama3-7B model, so that after the Llama3-7B model identifies the engineering event to which it belongs, it inserts the engineering text into the timeline of the engineering event to which it belongs according to the recording time of the engineering event according to the first fine-tuning instruction, integrates the timeline according to the second fine-tuning instruction to generate a timeline summary of the engineering event to which it belongs, and integrates all timeline summaries in the segment according to the third fine-tuning instruction to generate a segment summary.

4. The incremental timeline engineering information extraction method based on blockchain as claimed in claim 3 is characterized in that: Before inserting the project text into the timeline, first determine whether the current project text is related to the project event in the current timeline and whether the project text has available information. If the current project text is related to the project event in the current timeline and has available information, the current project text is inserted into the timeline, otherwise, it is not inserted.

5. The incremental timeline engineering information extraction method based on blockchain as claimed in claim 1 is characterized in that: The event line of each engineering event is stored in the storage space of the corresponding shard in the form of a list. When generating a summary, the list content of the corresponding engineering event is directly sorted to obtain a timeline summary of the corresponding engineering event.

6. The incremental timeline engineering information extraction method based on blockchain as claimed in claim 1 is characterized in that: Each shard contains at least members from all relevant parties of the project, and each member is assigned to the corresponding shard according to his or her role and relevance in the project.

7. The incremental timeline engineering information extraction method based on blockchain as claimed in claim 1 is characterized in that: When a transaction within a shard is completed, the resulting transaction data is broadcast to other shards through cross-shard synchronization operations, so that all nodes participating in the corresponding project on all shards can access the latest transaction data.

8. The incremental timeline engineering information extraction method based on blockchain as claimed in claim 1 is characterized in that: The trigger condition for updating the timeline summary is that the user submits the project text to the blockchain, the trigger condition for updating the shard summary is that the timeline summary is updated, and the global summary is updated according to the preset period.

9. The incremental timeline engineering information extraction method based on blockchain as claimed in claim 1, characterized in that: The engineering documents include monitoring reports or construction logs or material purchase lists or rectification orders.

10. An incremental timeline engineering information extraction system based on blockchain, characterized in that: It includes a blockchain and a smart contract deployed on the blockchain, wherein the blockchain includes multiple shards, and different shards are used to process different engineering projects. The smart contract includes a transaction shard contract, a transaction formation contract, and a transaction supplement contract, wherein: The transaction sharding contract is used to use the sharding classification model to identify the engineering project to which the engineering text uploaded by the user belongs and send the transaction task of the engineering text to the shard corresponding to the engineering project; The transaction formation contract is used to execute the transaction task when the shard receives the transaction task. Different shards execute their own transaction tasks in parallel. The transaction task includes generating a timeline summary and a shard summary. The process of generating the timeline summary includes: firstly using the event clustering model to identify the engineering theme to which the engineering text belongs in the corresponding engineering project, and then using the event classification model to identify the engineering event to which the engineering text belongs in the corresponding engineering theme, inserting the engineering text into the timeline of the engineering event to which it belongs according to the recording time of the engineering event, integrating the timeline to generate the timeline summary of the engineering event to which it belongs, and the timeline is obtained by sorting different description files for the same engineering event according to time; the process of generating the shard summary includes: integrating all the timeline summaries in the shard to generate the shard summary of the corresponding engineering project; The transaction supplement contract is used to integrate all shard summaries to generate a global summary; The fragmentation classification model, event clustering model and event classification model are all trained neural network models.

Images