Transaction method for transaction element recognition based on large language model and related device

Abstract

The invention discloses a transaction method for transaction element recognition based on a large language model and a related device, relates to the field of artificial intelligence or finance, and aims at generating transaction record information through a to-be-tested negotiation transaction intention text in a text dialogue window and avoiding the process of manually recording and arranging transaction elements. The inquiry transaction information is automatically received, the transaction elements in the inquiry transaction information and the transaction elements in the transaction record information are compared, the price return transaction information is automatically generated and sent, the transaction information is confirmed, frequent manual intervention is not needed, automation of the transaction process is achieved, the transaction can be achieved more quickly, and the transaction efficiency is improved. And the overall transaction efficiency is improved. The quotation party does not need to check the electronic transaction platform in real time, does not need to manually check and inquire the transaction elements in the transaction information, and does not need to manually input the transaction price, so that the conditions of recognition errors, checking errors and input errors of manual checking of the transaction elements are avoided, and the operation complexity is also reduced.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of artificial intelligence, and in particular to a transaction method for transaction element identification based on a large language model and related devices. BACKGROUND

[0002] In the process of financial transactions (such as foreign exchange transactions), both parties to the transaction (including the quoting party and the bidding party) need to reach a transaction in the text chat window of the electronic transaction platform, that is, both parties to the transaction can conduct text negotiation through the text chat window, and the negotiation text negotiated through the text chat window includes multiple transaction elements and a transaction price.

[0003] After determining that a transaction is reached through the text chat window, the quoting party needs to check whether there is bidding transaction information sent by the bidding party in the electronic transaction platform in real time; check whether there is transaction information sent by the bidding party in the electronic transaction platform in real time, resulting in cumbersome operation. SUMMARY

[0004] In view of the above problems, the present application provides a transaction method for transaction element identification based on a large language model and related devices to achieve the purpose of simplifying the transaction process. The specific scheme is as follows:

[0005] The first aspect of the present application provides a transaction method for transaction element identification based on a large language model, comprising:

[0006] Obtaining a to-be-tested negotiation transaction intention text in a text chat window, the to-be-tested negotiation transaction intention text including multiple transaction elements and a transaction price;

[0007] Generating transaction record information based on the to-be-tested negotiation transaction intention text, the transaction record information including the multiple transaction elements and the transaction price;

[0008] If the bidding transaction information including the transaction element is received, obtaining target transaction record information matching the transaction element in the bidding transaction information from the multiple transaction record information;

[0009] Generating a counter-offer transaction information containing the transaction price in the target transaction record information;

[0010] Sending the counter-offer transaction information to a first terminal held by the bidding party;

[0011] Receiving transaction information representing that the bidding party determines that the transaction price in the target transaction record information is correct;

[0012] If the transaction price in the transaction information is the same as the transaction price in the target transaction record information, generating confirmation transaction information.

[0013] In a possible implementation, the generating the transaction record information based on the to-be-tested negotiation transaction intention text comprises:

[0014] extracting the plurality of transaction elements and the transaction price from the to-be-tested negotiation transaction intention text;

[0015] controlling display of the transaction elements and the transaction price;

[0016] generating the transaction record information in response to confirmation of no error.

[0017] In a possible implementation, the extracting the plurality of transaction elements and the transaction price from the to-be-tested negotiation transaction intention text comprises:

[0018] inputting the to-be-tested negotiation transaction intention text into a pre-constructed transaction element extraction model, and obtaining, by the transaction element extraction model, the transaction elements and the transaction price in the to-be-tested negotiation transaction intention text.

[0019] In a possible implementation, the method for training the transaction element extraction model comprises:

[0020] inputting a simulated negotiation transaction intention text into the transaction element extraction model, and outputting, by the transaction element extraction model, an extraction element result, the extraction element result comprising a plurality of transaction elements and a transaction price;

[0021] obtaining difference information between a labeled result of the simulated negotiation transaction intention text and the extraction element result, the labeled result comprising a plurality of transaction elements and a transaction price;

[0022] training the transaction element extraction model based on the difference information.

[0023] In a possible implementation, the plurality of transaction elements comprises a transaction counterparty name, a transaction price, a transaction term, a transaction date, a transaction direction, and a transaction variety type, and the method for obtaining each simulated negotiation transaction intention text comprises:

[0024] randomly obtaining a target transaction counterparty name from a plurality of preset transaction counterparty names;

[0025] randomly obtaining an amount value within a preset amount data range;

[0026] determining that a preset amount identifier, the amount value, an amount unit, and a first preset interference character constitute a target transaction price;

[0027] randomly obtaining a term value within a preset term data range;

[0028] The time limit value, the time limit unit, and the second preset interference character constitute a target transaction date;

[0029] Randomly obtaining a pending transaction direction from a plurality of preset transaction directions;

[0030] The pending transaction direction and the third preset interference character constitute a target transaction direction;

[0031] Combining the target transaction counterparty name, the target transaction date, and the target transaction direction to obtain the simulated negotiation transaction intention text.

[0032] In a possible implementation, the transaction element extraction model comprises:

[0033] A first vocabulary mapping module, an input end of the first vocabulary mapping module being an input end of the transaction element extraction model;

[0034] A vocabulary embedding module, an input end of the vocabulary embedding module being connected with an output end of the first vocabulary mapping module; an output end of the vocabulary embedding module being connected with an input end of a concatenation module;

[0035] A position embedding module, an input end of the position embedding module being connected with an output end of the first vocabulary mapping module; an output end of the position embedding module being connected with an input end of the concatenation module;

[0036] The concatenation module, an output end of the concatenation module being connected with an input end of a first normalization module, an input end of a first gating mechanism module, and an input end of a filtering module respectively;

[0037] The first normalization module, an output end of the first normalization module being connected with an input end of a masked multi-head self-attention module;

[0038] The masked multi-head self-attention module, an output end of the masked multi-head self-attention module being connected with an input end of a first DropOut module;

[0039] The first DropOut module, an output end of the first DropOut module being connected with an input end of a first residual connection module and an output end of the first gating mechanism module respectively;

[0040] The first residual connection module, an output end of the first residual connection module being connected with an input end of a second normalization module and an input end of a second gating mechanism module respectively;

[0041] The second normalization module, an output end of the second normalization module being connected with an input end of an MLP feedforward network;

[0042] The MLP feedforward network, an output end of the MLP feedforward network is connected with an input end of a second DropOut module;

[0043] The second DropOut module, an output end of the second DropOut module is connected with an input end of a second residual connection module;

[0044] The second residual connection module, an input end of the second residual connection module is connected with an output end of the second gating mechanism module; an output end of the second residual connection module is connected with an input end of a weighted cross-layer connection module;

[0045] The weighted cross-layer connection module, an input end of the weighted cross-layer connection module is connected with an output end of the filtering module; a pure output end of the weighted cross-layer connection module is connected with an input end of a linear layer;

[0046] The linear layer, an output end of the linear layer is connected with an input end of a Softmax layer;

[0047] The Softmax layer, an output end of the Softmax layer is connected with an input end of a second vocabulary mapping module;

[0048] The second vocabulary mapping module, an output end of the second vocabulary mapping module is an output end of a transaction element extraction model;

[0049] The filtering module;

[0050] The first gating mechanism module;

[0051] The second gating mechanism module.

[0052] In a possible implementation, the first gating mechanism module or the second gating mechanism module comprises:

[0053] A first fully connected layer, an input end of the first fully connected layer is an input end of the first gating mechanism module or the second gating mechanism module;

[0054] A first activation function connected with an output end of the first fully connected layer, an output end of the first activation function is an output end of the first gating mechanism module or the second gating mechanism module.

[0055] In a possible implementation, the filtering module comprises:

[0056] A second fully connected layer, an input end of the second fully connected layer is an input end of the filtering module;

[0057] A second activation function connected with an output end of the second fully connected layer; an output end of the second activation function is an output end of the filtering module.

[0058] The second aspect of the present application provides a transaction device based on transaction element identification of a large language model, comprising:

[0059] A first acquisition module is configured to acquire a to-be-tested negotiation transaction intention text in a text dialogue window, wherein the to-be-tested negotiation transaction intention text comprises a plurality of transaction elements and a transaction price;

[0060] A first generation module is configured to generate transaction record information based on the to-be-tested negotiation transaction intention text, wherein the transaction record information comprises the plurality of transaction elements and the transaction price;

[0061] A second acquisition module is configured to acquire target transaction record information matching transaction elements in inquiry transaction information from the plurality of transaction record information if the inquiry transaction information including the transaction elements is received;

[0062] A second generation module is configured to generate counter-offer transaction information containing the transaction price in the target transaction record information;

[0063] A first sending module is configured to send the counter-offer transaction information to a first terminal held by an inquiring party;

[0064] A first receiving module is configured to receive transaction completion information representing that the inquiring party determines that the transaction price in the target transaction record information is correct;

[0065] A third generation module is configured to generate confirmation transaction completion information if the transaction price in the transaction completion information is the same as the transaction price in the target transaction record information.

[0066] The third aspect of the present application provides a computer program product comprising computer readable instructions, which, when executed on an electronic device, cause the electronic device to implement the transaction method based on transaction element identification of a large language model of the first aspect or any implementation manner of the first aspect.

[0067] The fourth aspect of the present application provides an electronic device comprising at least one processor and a memory connected to the processor, wherein:

[0068] The memory is configured to store a computer program;

[0069] The processor is configured to execute the computer program, so that the electronic device can implement the transaction method based on transaction element identification of a large language model of the first aspect or any implementation manner of the first aspect.

[0070] The fifth aspect of the present application provides a computer storage medium, the storage medium carries one or more computer programs, when the one or more computer programs are executed by an electronic device, the electronic device can implement the transaction method for transaction element identification based on a large language model of the first aspect or any implementation manner of the first aspect.

[0071] By the above technical solution, the present application provides a transaction method for transaction element identification based on a large language model, which generates transaction record information from the to-be-tested negotiation transaction intention text in the text dialogue window, avoids the process of manually recording and arranging transaction elements, and greatly saves time. The present application automatically receives inquiry transaction information, compares transaction elements in the inquiry transaction information and transaction elements in the transaction record information, automatically generates and sends back price transaction information, confirms the transaction information, and performs a series of operations without frequent human intervention, realizes the automation of the transaction process, enables the transaction to be achieved more quickly, and improves the overall transaction efficiency. The offeror does not need to check the electronic transaction platform in real time, does not need to manually check the transaction elements in the inquiry transaction information, and does not need to manually input the transaction price, thereby avoiding the situation of manual checking of transaction element identification errors, checking errors, and input errors. For the offeror, the transaction operation process is simplified, which enables it to focus more on the formulation of transaction strategies and the analysis of market dynamics, rather than wasting a lot of energy on tedious manual operations and information checking, thereby improving work efficiency and satisfaction. For the inquiry party and the offeror, the transaction process is more smooth and efficient, and the transaction intention can be achieved in time, improving the user experience of the electronic transaction platform. BRIEF DESCRIPTION OF DRAWINGS

[0072] The above and other features, advantages, and aspects of the present disclosure will become more apparent with reference to the following detailed description in conjunction with the accompanying drawings. Throughout the drawings, the same or similar reference numerals denote the same or similar elements. It should be understood that the drawings are schematic, and the original and elements are not necessarily drawn according to the scale.

[0073] Figure 1 A system architecture schematic diagram is provided for the present application;

[0074] Figure 2 A flowchart of a transaction method for transaction element identification based on a large language model is provided for the embodiment of the present application;

[0075] Figure 3 A generation process schematic diagram of transaction record information is provided for the embodiment of the present application;

[0076] Figure 4 A schematic diagram of an implementation manner of a transaction element extraction model is provided for the embodiment of the present application;

[0077] Figure 5 A structural schematic diagram of the first gating mechanism module or the second gating mechanism module provided by the embodiment of the present application is shown in FIG. 1.

[0078] Figure 6 A structural schematic diagram of the filtering module provided by the embodiment of the present application is shown in FIG. 2.

[0079] Figure 7 A structural schematic diagram of the transaction device based on the transaction element identification of the large language model provided by the embodiment of the present application is shown in FIG. 3.

[0080] Figure 8 A structural schematic diagram of the electronic device provided by the embodiment of the present application is shown in FIG. 4. DETAILED DESCRIPTION

[0081] The embodiments of the present application are described below in conjunction with the accompanying drawings. The terms used in the embodiment part of the present application are only used to explain the specific embodiments of the present application, and are not intended to limit the present application.

[0082] The embodiments of the present application are described below in conjunction with the accompanying drawings. It is known to those of ordinary skill in the art that, as technology develops and new scenarios appear, the technical solutions provided by the embodiments of the present application are also applicable to similar technical problems.

[0083] The terms “first”, “second”, and the like in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and do not necessarily indicate a specific order or sequence. It should be understood that the terms used in this way can be interchanged under appropriate circumstances, and this is only a distinguishing way used in the description of the embodiments of the present application to describe the objects with the same attributes. In addition, the terms “include” and “have” and any variations thereof are intended to cover non-exclusive inclusion, so that the processes, methods, systems, products or devices containing a series of units do not have to be limited to those units, but can include other units that are not clearly listed or inherent to these processes, methods, products or devices.

[0084] During the process of financial transactions (such as foreign exchange transactions), the transaction parties need to reach a transaction in the text chat window of the electronic transaction platform, that is, the transaction parties can conduct text negotiation through the text chat window, and the negotiation text negotiated through the text chat window includes but is not limited to transaction elements such as transaction direction, transaction counterparty name, transaction price, transaction quantity, and transaction variety information.

[0085] After determining the transaction through the text conversation window, in the related technology, the inquiring party artificially obtains the transaction elements from the negotiation transaction intention text, and generates inquiry transaction information containing the transaction elements through the first terminal held by the inquiring party; the inquiry transaction information is sent to the second terminal held by the offering party through the electronic transaction platform, and the offering party artificially checks whether the transaction elements in the inquiry transaction information are the same as the negotiation transaction intention text that has been determined, if they are the same, the offering party artificially inputs the transaction price in the specified field of the inquiry prompt page displayed on the second terminal, and manually clicks "counter-offer", so as to send the counter-offer transaction information containing the transaction price to the first terminal; the inquiring party artificially checks whether the transaction price in the counter-offer transaction is the same as the negotiation transaction intention text that has been determined, if they are the same, the transaction completion information can be sent to the second terminal; the offering party artificially determines the transaction completion information through the second terminal, so as to complete the transaction.

[0086] In the above related technology, the offering party needs to view in real time whether there is inquiry transaction information sent by the inquiring party in the electronic transaction platform; needs to view in real time whether there is confirmation transaction completion information sent by the inquiring party in the electronic transaction platform; and needs to artificially check whether the transaction elements in the inquiry transaction are the same as the transaction elements in the negotiation transaction intention text that has been determined. In summary, there are multiple artificial judgments and operations in the process of completing a negotiation transaction, there is a possibility of artificial operation error, and there is a problem of low artificial operation efficiency.

[0087] Based on this, the embodiment of the present application provides a transaction method based on transaction element recognition of a large language model. After the transaction between the two parties is reached through the text conversation window, transaction record information can be generated based on the negotiation transaction intention text to be tested in the text conversation window. The transaction record information includes transaction elements and a transaction price. After receiving the inquiry transaction information, the transaction elements in the inquiry transaction information are automatically matched with the transaction elements in the transaction record information. If the target transaction record information is matched, counter-offer transaction information containing the transaction price in the target transaction record information is automatically generated and sent to the first terminal held by the inquiring party. If the inquiring party confirms that there is no error, the first terminal can generate transaction completion information and send it to the second terminal held by the offering party. The second terminal automatically detects whether the transaction price in the transaction completion information is the same as the transaction price in the target transaction record information. If they are the same, confirmation transaction completion information is automatically generated and fed back to the first terminal. In the embodiment of the present application, the offering party does not need to view the electronic transaction platform in real time, and automatically completes each operation, avoiding the possibility of artificial operation and improving efficiency.

[0088] This application automatically receives inquiry transaction information, matches transaction elements, generates and sends counter-offer transaction information, and confirms transaction information, without requiring frequent manual intervention. This automates the transaction process, enabling transactions to be completed more quickly and improving overall transaction efficiency.

[0089] In related technologies, the transaction process is cumbersome, requiring manual monitoring of the electronic trading platform and multiple checks and operations, which easily leads to human error, such as misidentification of transaction elements, verification errors, and incorrect input of transaction prices, potentially resulting in transaction failure or economic losses. This application, however, utilizes automated processes and accurate identification and matching of transaction elements, reducing manual intervention, mitigating the risks caused by human error, and improving the accuracy and reliability of transactions.

[0090] For the quoting party, the simplified transaction process allows them to focus more on developing trading strategies and analyzing market dynamics, rather than wasting significant time on tedious manual operations and information verification, thus improving work efficiency and satisfaction. For both the requesting and quoting parties, the transaction process is smoother and more efficient, enabling timely reaching of transaction intentions and enhancing the user experience of the electronic trading platform.

[0091] The following describes the transaction method and related apparatus for identifying transaction elements based on a large language model, as provided in the embodiments of this application.

[0092] See Figure 1 , Figure 1 A schematic diagram of a system architecture is shown. The system may include a first terminal 100, a server 200, and a second terminal 300.

[0093] Among them, server 200 runs an electronic trading platform.

[0094] For example, the first terminal 100 is a terminal held by the party making the inquiry. The second terminal 300 is a terminal held by the party making the offer.

[0095] For example, the interfaces displayed through the first terminal 100 include, but are not limited to: text dialog window, inquiry transaction interface, and transaction interface.

[0096] For example, the interface displayed through the second terminal 300 includes, but is not limited to, a text dialog window.

[0097] Exemplarily, through the text conversation window displayed by the first terminal 100 and the text conversation window displayed by the second terminal 300, the inquiring party and the offering party can conduct a negotiated dealing market transaction, so as to obtain a to-be-tested negotiated transaction intention text, the to-be-tested negotiated transaction intention text including multiple transaction elements and a transaction price. After the offering party and the inquiring party successfully negotiate through the text conversation window, the first terminal 100 displays an inquiry transaction interface, so that the inquiring party manually inputs multiple transaction information through the inquiry transaction interface to generate inquiry transaction information, or automatically extracts the multiple transaction elements from the to-be-tested negotiated transaction intention text, automatically inputs the multiple transaction information on the inquiry transaction interface, and if the inquiring party confirms that the information is correct, the inquiry transaction information can be generated; the first terminal 100 sends the inquiry transaction information to the second terminal 300 through the server 200. The second terminal 300 can execute the transaction method for transaction element recognition based on a large language model provided in the embodiments of the present application, sends the counter-offer transaction information to the first terminal 100 through the server 200, and the first terminal 100 can display a transaction completion interface, the transaction completion interface displaying the counter-offer transaction information; if the inquiring party considers that the transaction price is correct, the transaction completion information can be generated and fed back to the second terminal 300 through the server 200, and the second terminal 300 can execute the transaction method for transaction element recognition based on a large language model provided in the embodiments of the present application.

[0098] Exemplarily, through the text conversation window displayed by the first terminal 100 and the text conversation window displayed by the second terminal 300, the inquiring party and the offering party can conduct a negotiated dealing market transaction, so as to obtain a to-be-tested negotiated transaction intention text, the to-be-tested negotiated transaction intention text including multiple transaction elements and a transaction price. After the offering party and the inquiring party successfully negotiate through the text conversation window, the first terminal 100 displays an inquiry transaction interface, so that the inquiring party manually inputs multiple transaction information through the inquiry transaction interface to generate inquiry transaction information, or automatically extracts the multiple transaction elements from the to-be-tested negotiated transaction intention text, automatically inputs the multiple transaction information on the inquiry transaction interface, and if the inquiring party confirms that the information is correct, the inquiry transaction information can be generated; the first terminal 100 sends the inquiry transaction information to the server 200. The server 200 can execute the transaction method for transaction element recognition based on a large language model provided in the embodiments of the present application, sends the counter-offer transaction information to the first terminal 100, and the first terminal 100 can display a transaction completion interface, the transaction completion interface displaying the counter-offer transaction information; if the inquiring party considers that the transaction price is correct, the transaction completion information can be generated and sent to the server 200, and the server 200 can execute the transaction method for transaction element recognition based on a large language model provided in the embodiments of the present application.

[0099] In summary, the transaction method based on the transaction element recognition of the large language model provided in the embodiments of the present application can be applied to the second terminal or the server.

[0100] For example, the algorithm for automatically extracting multiple transaction elements and a transaction price from the text of the negotiation transaction intention to be tested can be a transaction element extraction model, which can be located in the server 200 or the first terminal or the second terminal.

[0101] Next, the product form of the second terminal 300 is described. Figure 1

[0102] The second terminal 300 in the embodiments of the present application can be a mobile phone, a tablet computer, a wearable device, a vehicle-mounted device, an augmented reality (AR) / virtual reality (VR) device, a notebook computer, an ultra-mobile personal computer (UMPC), a netbook, a personal digital assistant (PDA), and the like, and the embodiments of the present application do not make any limitation thereto.

[0103] The second terminal 300 can include a radio frequency unit, a memory, an input unit, a display unit, a camera (optional), an audio circuit (optional), a speaker (optional), a microphone (optional), an earphone jack (optional), a processor, an external interface, a power supply, and the like. Those skilled in the art can understand that the above-mentioned components are only examples and do not constitute a limitation on the terminal or the multi-functional device, and can include more or fewer components, or combine certain components, or different components.

[0104] The input unit can be used to receive inputted digital or character information, and to generate key signal input related to user settings and function control of the portable multi-functional device. Specifically, the input unit can include a touch screen (optional) and / or other input devices. Specifically, the other input devices can include, but are not limited to, one or more of a physical keyboard, function keys (such as volume control buttons, on-off buttons, etc.), trackballs, mice, joysticks, and the like.

[0105] Among them, the input device can receive inputted data and the like.

[0106] The display unit can be used to display information inputted by the user or provided to the user, various menus of the terminal, interactive interfaces, file display, and / or playing of any kind of multimedia files. In the embodiments of the present application, the display unit can be used to display interfaces, processing results, and the like.

[0107] ​The memory can be configured to store software code related to the transaction method based on the large language model transaction element identification, the processor can execute the steps of the transaction method based on the large language model transaction element identification, and can also dispatch other units (such as the input unit and the display unit) to realize the corresponding functions.

[0108] The radio frequency unit (optional) can be used for receiving and sending signals in the process of information or communication.

[0109] In the embodiments of the present application, the radio frequency unit can send data to the server 200 and receive the processing result sent by the server 200.

[0110] It should be understood that the radio frequency unit is optional and can be replaced by other communication interfaces, such as a network interface.

[0111] The second terminal 300 further includes a power supply (such as a battery) for supplying power to each component.

[0112] The second terminal 300 further includes an external interface, which can be a standard Micro USB interface or a multi-pin connector, and can be used for connecting the second terminal 300 with other devices for communication or connecting a charger for charging the second terminal 300.

[0113] The server 200 includes a bus, a processor, a communication interface and a memory. The processor, the memory and the communication interface communicate through the bus.

[0114] The memory can be configured to store software code related to the transaction method based on the large language model transaction element identification, the processor can execute the steps of the transaction method based on the large language model transaction element identification, and can also dispatch other units to realize the corresponding functions.

[0115] Reference Figure 2 , Figure 2 A flowchart of a transaction method based on large language model transaction element identification provided by the embodiments of the present application is shown in Figure 2 The transaction method based on large language model transaction element identification provided by the embodiments of the present application can include steps S201 to S207, which will be described in detail below.

[0116] Step S201: Obtain the to-be-tested negotiation transaction intention text in the text dialogue window, wherein the to-be-tested negotiation transaction intention text includes multiple transaction elements and a transaction price.

[0117] Some transactions are made through the text chat window of the electronic trading platform, which are usually referred to as negotiated dealing market transactions. The making of a negotiated dealing transaction usually needs to comply with the specified transaction confirmation logic, that is, the transaction parties output and confirm the text of multiple transaction elements and the transaction price through the stages of inquiry, offer, counteroffer, and acceptance, so as to form the negotiated dealing intention text to be tested.

[0118] For example, the negotiated dealing intention text to be tested includes the transaction dialogue between the buyer and the seller during the transaction negotiation.

[0119] Business negotiation refers to the process in which the buyer and the seller negotiate the making of a transaction through direct communication or letter, so as to complete the transaction. The transaction made in this way is referred to as the negotiated dealing intention before it is actually recorded in the transaction account books of the parties.

[0120] For example, the multiple transaction elements include but are not limited to the name of the counterparty, the transaction direction, the transaction period, and the transaction type.

[0121] The name of the counterparty refers to the formal name or identity of the other party in the financial transaction.

[0122] The transaction direction refers to the specific buying or selling operation direction in the financial transaction. In different transactions, the transaction direction can be different or the same. For example, in the foreign exchange transaction, the transaction direction of the inquiring party includes but is not limited to “buy” (buying) and “sell” (selling). In the foreign exchange swap transaction, the transaction direction of the inquiring party includes but is not limited to “buy / sell” (near-end buying and far-end selling) and “sell / buy” (near-end selling and far-end buying). The transaction direction of the offering party includes but is not limited to “Bid” and “Offer”.

[0123] The transaction period refers to the time range agreed by the transaction parties for completing the transaction or fulfilling the contract obligations, which clearly indicates the length of time or the specific delivery date required for the transaction from making to completion, and is an important element affecting the transaction risk, income, and capital occupation. For example, in the foreign exchange spot transaction, the transaction period is usually two business days (T+2) for delivery; while in the foreign exchange forward transaction, the transaction period can be a specific future date agreed by the transaction parties, such as 1 month, 3 months, or 1 year, for delivery.

[0124] The transaction variety type refers to the product type of the specific product of the transaction. In foreign exchange transactions, the transaction variety type can include but is not limited to: transaction currency (for example, "first currency / second currency").

[0125] The transaction price is the actual transaction price reached by the buyer and the seller in the market.

[0126] The following is an example of foreign exchange transactions to illustrate the to-be-tested negotiation intention text. Assume that the to-be-tested negotiation intention text is "We b / s first currency / second currency 6s in -1020".

[0127] Among them, "We" refers to the name of the counterparty. "b / s" refers to the transaction direction, "b / s" is the abbreviation of "buy / sell". "First currency / second currency" refers to the transaction variety type, indicating that "first currency" is spent to buy "second currency", that is, Buy first currency Sell second currency. "6s" refers to the transaction period, which is 6 months here. The transaction period can be expressed in full or in abbreviation. "in" is used to connect with the subsequent transaction price to represent the transaction price; "-1020" refers to the transaction price.

[0128] For example, "in" can be replaced by words such as 'at' and 'on'.

[0129] For example, the to-be-tested negotiation intention text can or can not include the name of the counterparty. For example, the name of the counterparty can be replaced by the name of the transaction subject.

[0130] For example, the to-be-tested negotiation intention text can or can not include the transaction direction. For example, if the to-be-tested negotiation intention text includes the transaction direction, the to-be-tested negotiation intention text should include the transaction variety type.

[0131] Step S202: generating transaction record information based on the to-be-tested negotiation intention text, the transaction record information including the plurality of transaction elements and the transaction price.

[0132] For example, the transaction record information can also include the element type to which the transaction element belongs and the element type to which the transaction price belongs.

[0133] The following illustrates the transaction record information by taking the transaction record information stored in the form of a table as an example. See Table 1.

[0134] Table 1

[0135]

[0136] Each row in Table 1 is a transaction record information.

[0137] As shown in Table 1, the specific value "-1234" of the transaction price belongs to the element type of transaction price; the specific value "first currency / second currency" of the transaction type belongs to the element type of transaction type, and other element types can be referred to Table 1, which will not be described here.

[0138] For example, a transaction number can be assigned to each transaction record information, so as to distinguish different transaction record information.

[0139] For example, the steps S201 to S207 can be applied to the second terminal or server held by the offeror, so that the second terminal or server can receive the inquiry transaction information sent by the first terminal. The transaction direction corresponding to the offeror is ask (inquiry), so the transaction direction in the transaction record information refers to the transaction direction corresponding to the offeror, rather than the transaction direction corresponding to the bidder, so as to facilitate the subsequent automatic matching process. For example, the transaction counter name in the transaction record information is the name of the offeror.

[0140] For example, the inquiry transaction information can be matched with the transaction record information. It can be understood that the transaction direction of the offeror and the transaction direction of the bidder are opposite and also belong to a kind of matching; therefore, the transaction direction in the transaction record information can be the transaction direction corresponding to the offeror.

[0141] For example, the transaction record information can be stored in the memory list.

[0142] Step S203: If the inquiry transaction information including the transaction elements is received, the target transaction record information matched with the transaction elements in the inquiry transaction information is obtained from the plurality of transaction record information.

[0143] For example, the plurality of transaction elements in the inquiry transaction information can be input by the bidder.

[0144] For example, the target transaction record information matched with the transaction elements in the inquiry transaction information can be the target transaction record information containing the transaction elements in the inquiry transaction information.

[0145] Step S204: The counter offer transaction information containing the transaction price in the target transaction record information is generated.

[0146] For example, the counter offer transaction information is the counter offer process of the offeror.

[0147] In the related art, the inquiring party submits the inquiry transaction information at an indefinite time, and the offering party needs to handle other work during this period, which may not be able to find the inquiry transaction information submitted by the inquiring party in time, resulting in that the inquiry transaction information is not handled in time. Even if the inquiry transaction information is found, the transaction elements are manually checked. The embodiment of the present application automatically receives the inquiry transaction information, automatically matches the inquiry transaction information with the transaction record information, and automatically generates the counter-offer transaction information, without the need for the offering party to wait for the inquiring party to submit the inquiry transaction information using the electronic transaction platform. There is no need for manual checking of transaction elements, which avoids the checking error that may occur when manually checking transaction elements.

[0148] Step S205: sending the counter-offer transaction information to the first terminal held by the inquiring party.

[0149] In the related art, the offering party needs to manually input the transaction price, while in the present application, the counter-offer transaction information containing the transaction price is automatically generated. This avoids the situation of manually inputting the transaction price by human error.

[0150] Step S206: receiving the transaction information representing that the inquiring party determines that the transaction price in the target transaction record information is correct.

[0151] It can be understood that if the inquiring party confirms that the transaction price is correct, the first terminal can generate the transaction information and send it to the second terminal.

[0152] Step S207: if the transaction price in the transaction information is the same as the transaction price in the target transaction record information, generate the confirmation transaction information.

[0153] In the related art, the offering party needs to wait for the transaction price confirmed by the inquiring party through the electronic transaction platform, and click the transaction intention, which may complete the transaction. If there is a human error operation, it may lead to the rejection of the transaction, resulting in the failure of the transaction; while in the present application, in the case that the transaction price in the transaction information is the same as the transaction price in the transaction record information, the confirmation transaction information can be automatically confirmed, thereby avoiding the situation of operation error.

[0154] Exemplarily, the confirmation transaction information can be sent to the first terminal.

[0155] Exemplarily, after the transaction is completed, the target transaction record information can be deleted.

[0156] The embodiment of the application provides a transaction method based on transaction element identification of a large language model. The transaction record information is generated from the to-be-tested negotiation transaction intention text in the text dialogue window, which avoids the process of manually recording and arranging transaction elements, greatly saves time, and automatically receives the inquiry transaction information, compares the transaction elements in the inquiry transaction information and the transaction elements in the transaction record information, automatically generates and sends the counter-offer transaction information and the confirmation of the transaction information, and the like, without frequent manual intervention, realizes the automation of the transaction process, enables the transaction to be achieved more quickly, and improves the overall transaction efficiency. The offeror does not need to check the electronic transaction platform in real time, does not need to manually check the transaction elements in the inquiry transaction information, and does not need to manually input the transaction price, thereby avoiding the situation of manual checking of transaction element identification errors, checking errors and input errors. For the offeror, the transaction operation process is simplified, which enables the offeror to focus more on the formulation of a transaction strategy and the analysis of market dynamics, rather than wasting a lot of energy on tedious manual operation and information checking, thereby improving work efficiency and satisfaction. For the inquiry party and the offeror, the transaction process is more smooth and efficient, the transaction intention can be achieved in time, and the user experience of the electronic transaction platform is improved.

[0157] It can be understood that there are various implementation manners of step S202, and the embodiment of the application provides but is not limited to the following method, which includes the following steps A1 to A3 in the implementation process.

[0158] Step A1: extracting the plurality of transaction elements and the transaction price from the to-be-tested negotiation transaction intention text.

[0159] For example, the plurality of transaction elements and the transaction price can be manually extracted from the to-be-tested negotiation transaction intention text.

[0160] For example, the plurality of transaction elements and the transaction price can be automatically extracted from the to-be-tested negotiation transaction intention text.

[0161] It can be understood that manual identification of the transaction elements in the to-be-tested negotiation transaction intention text may appear identification errors, and if the transaction elements in the to-be-tested negotiation transaction intention text are automatically identified, the situation is avoided.

[0162] Step A2: controlling display of the transaction elements and the transaction price.

[0163] Step A3: in response to an operation of confirming no error, generating the transaction record information.

[0164] In order for those skilled in the art to better understand steps A1 to A3, the following illustrates the process.

[0165] As Figure 3As shown, a schematic diagram of the generation process of the transaction record information provided by the embodiment of the present application.

[0166] Figure 3 In (a), the original text entry window, the offeror can copy the to-be-tested negotiation intention text "B / S with ABCDE 10mio 9s -1234" to the to-be-recognized area 31, and click the "recognize" button, and the transaction elements and the transaction price can be displayed in the Figure 3 In (b), the negotiation intention transaction element extraction element result confirmation window shown in (b) displays the transaction elements and the transaction price. If the offeror clicks the "confirm" button, the transaction record information can be displayed in the Figure 3 In (c), the negotiation intention record and display window shown in (c) displays the transaction record information.

[0167] For example, the transaction record information can be written into the memory list.

[0168] It can be understood that there are many implementation manners of step A1, and the embodiments of the present application provide but are not limited to the following two.

[0169] The first implementation manner of step A1 includes the following step A11: extracting the plurality of transaction elements and the transaction price from the to-be-tested negotiation intention text based on a regular expression.

[0170] The to-be-tested negotiation intention text refers to a natural language text containing transaction information.

[0171] The regular expression (Regular Expression) is a text processing tool used to describe a pattern of a group of strings. In the implementation of a programming language, the regular expression uses a single string to describe and match a series of strings conforming to a certain syntax rule.

[0172] It can be understood that there are many negotiation intention texts at present, which are referred to as original texts in the embodiments of the present application. Different regular expressions are applied to the transaction direction, the transaction counterparty name, the transaction quantity, the transaction price, and the transaction variety type contained in the original text.

[0173] The second implementation manner of step A1 includes the following step A21: inputting the to-be-tested negotiation intention text into a pre-constructed transaction element extraction model, and acquiring the transaction elements and the transaction price in the to-be-tested negotiation intention text through the transaction element extraction model.

[0174] The transaction element extraction model provided in the application has fewer format requirements for the to-be-tested negotiation transaction intention text, and the transaction element extraction model has stronger generalization capability for different formats of the to-be-tested negotiation transaction intention text. The generalization capability of the model refers to the capability of a model to correctly understand and output prediction results when facing new and unseen data. That is, the model can perform well not only on training data but also on new and unknown data sets. Regular expressions need to limit the character length of each extracted transaction element. In the scenario applied in the application, the to-be-tested negotiation transaction intention text is a non-standard format transaction intention single, and in the process of manually inputting transaction elements, the input transaction elements may be missing necessary delimiters. For such cases, transaction element recognition based on regular expressions may not be able to recognize or may recognize incorrectly. However, transaction element recognition based on the transaction element extraction model can recognize such cases and output the corrected results.

[0175] The method for training the transaction element extraction model is described below, and the method for training the transaction element extraction model includes the following steps B1 to B3.

[0176] Step B1: input the simulated negotiation transaction intention text into the transaction element extraction model, and output the extracted element result through the transaction element extraction model, wherein the extracted element result includes multiple transaction elements and a transaction price.

[0177] The applicant considered the following content in selecting the transaction element extraction model:

[0178] The machine learning model has the following limitations: traditional machine learning models such as logistic regression and support vector machines are usually suitable for supervised learning tasks, and the labels of the training data are often category symbols, and these category symbols are usually within a fixed value range. For example, in a binary classification problem, the label can only be 0 or 1; in a multi-classification problem, the label can be 0, 1, 2, etc. limited several values. In the transaction element extraction scenario, the label of the training data is the transaction element extracted from the simulated negotiation transaction intention text, and the value range of these transaction elements is not fixed. For example, the transaction counterparty name can be any institution or individual name; the transaction price can be any numerical value; the transaction period can be various time expressions; and the transaction quantity can also vary within a large range. This label feature without a fixed value range makes it difficult to directly apply traditional machine learning models.

[0179] Traditional machine learning models usually have fixed requirements for the format of training data, often requiring input data to be converted into a specific vector form, and the number and order of features need to remain consistent during training and prediction. For example, the logistic regression model requires the input features to be a numerical vector, and the feature dimension of each sample is the same. However, in the transaction element extraction application scenario, the format and content of the simulated negotiation transaction intention text are very flexible. Different simulated negotiation transaction intention texts may contain different numbers and orders of transaction elements, and the natural language expression is diverse, and the simulated negotiation transaction intention text is not fixed. This makes it difficult for traditional machine learning models to handle such non-fixed format data.

[0180] Generative neural network models, especially those based on the Transformer architecture, can handle labels with no fixed value range very well. It can generate the corresponding transaction elements according to the content of the simulated negotiation transaction intention text, without needing to limit the value range of the label in advance. For example, as long as the Transformer-based model learns the relevant patterns and rules during training, it can generate names of any length and content for the transaction counterpart name, and any numerical amount for the transaction price.

[0181] Generative neural network models have strong generalization ability when processing natural language text. It can learn the deep semantics and grammatical structure of language, and thus understand and process simulated negotiation transaction intention texts of different formats and expressions. Even if the order of the simulated negotiation transaction intention text changes or contains some new expressions, a well-trained generative neural network model can still adapt and extract transaction elements well. This is because the model has learned a large number of different forms of simulated negotiation transaction intention texts during training, and can capture various forms of transaction elements in simulated negotiation transaction intention texts, thereby having the ability to cope with changing input texts.

[0182] Generative neural network models can naturally handle variable-length input and output sequences. Unlike traditional machine learning models, it does not need to perform strict fixed format conversion on input data, and can flexibly receive transaction texts of different lengths and structures, and generate corresponding transaction element extraction results. This strong adaptability to non-fixed format data makes the generative model have obvious advantages in handling complex natural language processing tasks.

[0183] Large language models are based on the classic Transformer model.

[0184] For example, the transaction element extraction model can be a generative neural network model or a large language model.

[0185] Step B2: Obtain the difference information of the labeling result of the simulated negotiation transaction intention text and the extracted element result, and the labeling result includes multiple transaction elements and a transaction price.

[0186] Step B3: Train the transaction element extraction model based on the difference information.

[0187] Through experimental verification, the accuracy rate of the transaction element extraction model in extracting transaction elements and transaction prices of the simulated negotiation transaction intention text is greater than 99.98%.

[0188] For example, the transaction element extraction model can be trained using an A100 GPU (Graphics Processing Unit).

[0189] It can be understood that the transaction parties can negotiate through a text chat window. Although the content of the to-be-tested negotiation transaction intention text obtained through the text chat window is roughly the same, the order and format of the multiple transaction elements contained in the to-be-tested negotiation transaction intention text have no strict requirements. The to-be-tested negotiation transaction intention text is based on the text chat window and cannot be automatically entered into an electronic transaction platform, i.e., cannot be automatically transacted. The inquiring party needs to initiate a request for quote (RFQ) transaction information that is completely consistent with the transaction elements in the to-be-tested negotiation transaction intention text to the quoting party through the RFQ function of the electronic transaction platform, and then the quoting party confirms the transaction of the RFQ transaction information.

[0190] It can be understood that the order and format of each transaction element in the to-be-tested negotiation transaction intention text are not uniform. The to-be-tested negotiation transaction intention text can also include the content of several natural language dialogues. There can also be interference characters in the transaction elements. This leads to a more difficult process of training the transaction element extraction model. A model based on the Transformer architecture requires a large amount of training data, i.e., simulated negotiation transaction intention texts. In the present application, it is difficult to obtain simulated negotiation transaction intention texts, and the labeling result of the simulated negotiation transaction intention text needs to be manually labeled, resulting in high cost and lack of feasibility. In order to train an accurate transaction element extraction model, the present application provides a method for obtaining simulated negotiation transaction intention texts and labeling results of simulated negotiation transaction intention texts.

[0191] It can be understood that the multiple transaction elements include a transaction counterparty name, a transaction price, a transaction term, a transaction date, a transaction direction, and a transaction variety type.

[0192] First, the characteristics of each transaction element need to be determined. The following describes an example of foreign exchange transactions.

[0193] The applicant disassembles the original text in the text conversation window, and maps the character interval of the original text as: 26 English letters, 0-9 digits, / , +, -, space. The counterparty name is a 4-character string composed of English letters; the transaction price is an indefinite-length string composed of numbers, spaces, and currency units; the transaction period is an indefinite-length string composed of numbers, period units, or pure period English names; the transaction price is an indefinite-length string composed of "+" or "-" and numbers; the transaction direction is an indefinite-length string composed of buy / sell English identifiers; and the transaction quantity is an indefinite-length string composed of numbers, spaces, and quantity units. For example, the order of multiple transaction elements in different original texts is not fixed. Based on this, the method for obtaining each simulated negotiation transaction intention text is described, which includes the following steps C1 to C8.

[0194] Step C1: Randomly obtain a target counterparty name from a plurality of preset counterparty names.

[0195] For example, the list of preset counterparty names is ["ABC Bank", "DEF Corp", "GHI Ltd"], and "DEF Corp" is randomly obtained as the target counterparty name.

[0196] Step C2: Randomly obtain an amount value within a preset amount data range.

[0197] For example, the preset amount data range is [0, 100 million], and a value within the range is randomly generated as the amount value, such as 567 million as the amount value.

[0198] Step C3: Determine a preset amount identifier, the amount value, an amount unit, and a first preset interference character to form a target transaction price.

[0199] For example, the preset amount identifier can be "+" or "-".

[0200] For example, the amount unit can be "million", "million", "hundred million", or "billion".

[0201] For example, the first interference character includes but is not limited to space, punctuation, separator, and special character.

[0202] For example, the target transaction price can contain one or more first interference characters, and the position of the first interference character in the target transaction price is random.

[0203] Step C4: Randomly obtain a period value within a preset period data range.

[0204] Exemplarily, the preset deadline data range is [1 month, 12 month], and a value in the range is randomly generated as the deadline value.

[0205] Step C5: determining the target transaction date composed of the deadline value, the deadline unit, and the second preset interference character.

[0206] Exemplarily, the deadline unit can be “day”, “week”, “month”, or “year”.

[0207] Exemplarily, the second preset interference character can be space, punctuation, separator, or special character.

[0208] Exemplarily, the target transaction date can contain one or more second interference characters, and the positions of the second interference characters in the target transaction date are random.

[0209] Step C6: randomly obtaining the pending transaction direction from a preset plurality of transaction directions.

[0210] Suppose the preset plurality of transaction directions are “buy”, “sell”, “b / s”, “s / b”, etc., and one of them is randomly selected as the pending transaction direction.

[0211] Step C7: determining the target transaction direction composed of the pending transaction direction and the third preset interference character.

[0212] Exemplarily, the third preset interference character can be space, punctuation, separator, or special character.

[0213] Step C8: combining the target transaction counterparty name, the target transaction price, the target transaction date, and the target transaction direction to obtain the simulated negotiation transaction intention text.

[0214] Exemplarily, the order of the target transaction counterparty name, the target transaction price, the target transaction date, and the target transaction direction in the simulated negotiation transaction intention text can be random.

[0215] Exemplarily, the simulated negotiation transaction intention text can not include the transaction counterparty name and / or the transaction direction.

[0216] Exemplarily, a sample data generation template can be preset. The sample data generation template contains a database of the above-mentioned transaction elements and interference characters, a program for extracting transaction elements from the database, and a program for combining each transaction element. Exemplarily, the database contains a set of values of each transaction element, and the set of values can be increased or decreased to adapt to changing transaction scenarios.

[0217] A large number of simulated negotiation transaction intention texts are generated through the above method. These simulated negotiation transaction intention texts have diversity in format, content and expression, and can cover various possible transaction scenarios and text variations, so that the transaction element extraction model is exposed to a rich variety of data samples in the training process, thereby improving the generalization ability of the transaction element extraction model for different types and formats of transaction texts, and better coping with various changes of the to-be-tested negotiation transaction intention texts in real scenarios.

[0218] Natural language expression has flexibility and uncertainty, and the same transaction element can be expressed in many different ways. The simulated negotiation transaction intention text can simulate various common natural language variations by introducing interference characters (such as spaces, punctuation, etc.) and different combinations, so that the transaction element extraction model learns to accurately extract transaction elements in complex text environments, and improves the adaptability and robustness of the transaction element extraction model to natural language variations.

[0219] In actual scenarios, the annotated simulated negotiation transaction intention text may be scarce and have a high acquisition cost. Through the above method, a large number of simulated negotiation transaction intention texts can be quickly and batch generated, effectively solving the data scarcity problem, providing sufficient data support for model training, and helping the model to better learn the characteristics and patterns of transaction elements, and improving the performance and accuracy of the model.

[0220] It can be understood that since the simulated negotiation transaction intention text is generated, the annotation result of the simulated negotiation transaction intention text can be easily obtained. For example, the transaction elements of the generated simulated negotiation transaction intention text without interference characters are spliced to obtain the annotation result.

[0221] For example, a plurality of simulated negotiation transaction intention texts can be divided into a training set and a test set; for example, a hash algorithm can be used to ensure that there is no repeated simulated negotiation transaction intention text between the training set and the test set.

[0222] It can be understood that for the negotiation transaction scenarios of other financial products, the simulated negotiation transaction intention texts in the negotiation transaction scenarios of other financial products can also be generated by replacing the interval ranges of the transaction elements such as the transaction product, the transaction counterparty and the transaction price.

[0223] In an optional implementation manner, as shown in Figure 4 FIG. 1 is a schematic diagram of an implementation manner of a transaction element extraction model provided by an embodiment of the present application, which includes the following modules:

[0224] A first vocabulary mapping module 401, an input end of the first vocabulary mapping module being an input end of the transaction element extraction model.

[0225] Figure 4 In the formula, "w_1 w_2 w_3 … w_n" represents n characters in the to-be-tested negotiation transaction intention text, and w_i represents a character located at the i-th position.

[0226] For example, the to-be-tested negotiation transaction intention text is an English text; the first vocabulary mapping module 401 is configured to convert the to-be-tested negotiation transaction intention text into capital characters and map the capital characters to ASCII code representations to obtain a target vector.

[0227] For example, since the to-be-tested negotiation transaction intention text is indefinite in length, the maximum character length of the to-be-tested negotiation transaction intention text is limited according to actual scene data observation.

[0228] For example, if the maximum character length is preset, if the length of the target vector is less than the maximum character length, the preset placeholder can be supplemented; if the length of the target vector is greater than the maximum character length, the target vector can be truncated. Thus, the vector input to the first vocabulary mapping module 401 is a vector of the maximum character length.

[0229] For example, in the embodiment of the present application, the transaction elements in the text dialogue window are all English and not Chinese. The case of English characters does not affect the text meaning of the transaction elements. In order to reduce the scale of the training data of the transaction element extraction model, the input characters of the transaction element extraction model are all set to be in capital. Based on this, the first vocabulary mapping layer has the function of converting the characters in the to-be-tested negotiation transaction intention text to capital. Thus, the situation that the training efficiency of the transaction element extraction model is reduced due to too large training data is avoided, and the problem of overfitting due to data skew is avoided.

[0230] The vocabulary embedding module 402 is connected to the output end of the first vocabulary mapping module; and the output end of the vocabulary embedding module is connected to the input end of the splicing module.

[0231] For example, the target vector includes a vector corresponding to each character in the to-be-tested negotiation transaction intention text; the vocabulary embedding module 402 is configured to perform embedding operation on the target vector, so as to extract a semantic vector of the to-be-tested negotiation transaction intention text and convert it into a word embedding vector, for example, Figure 4 In the formula, "TE_1 TE_2 TE_3 …TE_n" represents a word embedding vector, and TE_i represents a semantic vector of a character at the i-th position.

[0232] The position embedding module 403 is connected to the output end of the first vocabulary mapping module; and the output end of the position embedding module is connected to the input end of the splicing module.

[0233] For example, the position embedding module can encode the position of each character in the to-be-tested negotiation intention text in the to-be-tested negotiation intention text, thereby generating a position embedding vector, for example, Figure 4 In the middle, "PE_1 PE_2 PE_3…PE_n" represents the position embedding vector, where PE_i represents the position vector of the character at the i-th position.

[0234] The position embedding vector can capture the position information of each character in the to-be-tested negotiation intention text and the relative position information of each character with other characters.

[0235] The splicing module 404, the input end of the first normalization module, the input end of the first gating mechanism module and the input end of the filtering module are connected respectively.

[0236] For example, the splicing module can splice the word embedding vector and the position embedding vector to form a vector sequence that integrates word semantics and position information.

[0237] The first gating mechanism module 418 is used for controlling information flow through a gating mechanism and automatically adjusting the strength of the residual connection according to the characteristics of the input data. The first gating mechanism module outputs a processed vector sequence.

[0238] The first normalization module 405 is connected to the input end of the masked multi-head self-attention module.

[0239] For example, the first normalization module normalizes the vector sequence to obtain a first normalized vector, so that it has a more stable numerical range.

[0240] The masked multi-head self-attention module 406 is connected to the input end of the first DropOut module.

[0241] For example, the masked multi-head self-attention module uses a multi-head attention mechanism to calculate the relationship between different positions in the first normalized vector, while masking future information to maintain the autoregressive characteristics of sequence generation. The masked multi-head self-attention module outputs an updated vector sequence.

[0242] The first DropOut module 407 is connected to the input end of the first residual connection module and the output end of the first gating mechanism module.

[0243] For example, the first DropOut module randomly discards the output of a part of neurons to prevent overfitting. The first DropOut outputs a processed vector sequence.

[0244] The first residual connection module 408 is connected with the input end of the second normalization module and the input end of the second gating mechanism module.

[0245] The first residual connection module adds the processed vector sequence output by the first DropOut module and the processed vector sequence output by the first gating mechanism module to form a residual connection, which helps the gradient flow. The first residual connection module outputs a residual connection vector sequence.

[0246] The second gating mechanism module 419 controls the information flow through the gating mechanism and automatically adjusts the strength of the residual connection according to the characteristics of the residual connection vector sequence output by the first residual connection module. The second gating mechanism module outputs a processed vector sequence.

[0247] The second normalization module 409 is connected with the input end of the MLP feedforward network.

[0248] The second normalization module normalizes the residual connection vector sequence to obtain a second normalized vector.

[0249] The MLP (Multi - layer Perceptron, multi-layer perceptron) feedforward network 410 is connected with the input end of the second DropOut module.

[0250] The MLP feedforward network performs nonlinear transformation on the second normalized vector through a multi-layer perceptron. The MLP feedforward network outputs a changed vector sequence.

[0251] The second DropOut module 411 is connected with the input end of the second residual connection module.

[0252] The second DropOut module randomly discards the output of a part of neurons to prevent overfitting. The second DropOut module processes the changed vector sequence output by the MLP feedforward network to obtain a processed vector sequence.

[0253] The second residual connection module 412 is connected with the output end of the second gating mechanism module and the input end of the weighted cross-layer connection module.

[0254] The second residual connection module adds the processed vector sequence output by the second gating mechanism module and the processed vector sequence output by the second DropOut module to form a residual connection. The second residual connection module outputs a residual connection vector sequence.

[0255] a filtering module 417, configured to perform a filtering operation on the vector sequence output by the concatenation module, and extract useful information. The filtering module outputs a filtered vector sequence.

[0256] The weighted cross-layer connection module 413 is connected to the output end of the filtering module; and the output end of the weighted cross-layer connection module is connected to the input end of the linear layer.

[0257] For example, the weighted cross-layer connection module performs weighted connection between the vector sequence output by the concatenation module and the residual connection vector sequence output by the second residual connection module, and enhances the reference ability of the transaction element extraction model to the input vector. The weighted cross-layer connection module outputs a cross-layer connection vector sequence.

[0258] The linear layer 414 is connected to the input end of the Softmax layer.

[0259] For example, the linear layer is configured to perform linear transformation on the cross-layer connection vector sequence to obtain a linearly changed vector sequence.

[0260] For example, the transaction element extraction model provided by the embodiment of the present application uses a Decoder-Only structure. Compared with an Encoder-Decoder structure, the Decoder-Only structure simplifies the design of the transaction element extraction model, reduces the complexity of the transaction element extraction model, improves the calculation efficiency, and guarantees good model effect in the scenario of the present application. For example, the Decoder-Only structure used by the embodiment of the present application further includes a first layer normalization module, a second layer normalization module, a masked multi-head self-attention module, a first residual connection module, a second residual connection module, and an MLP feedforward network.

[0261] It can be understood that the linear layer can receive the vector output by the Decoder-Only structure.

[0262] The Softmax layer 415 is connected to the input end of the second vocabulary mapping module.

[0263] For example, the Softmax layer is configured to convert the linearly changed vector sequence into a probability distribution. The Softmax layer outputs a vector sequence representing a label probability distribution of different transaction elements.

[0264] The second vocabulary mapping module 416 is an output end of the transaction element extraction model.

[0265] For example, the second vocabulary mapping module is used to map the vector sequence of label probability distributions to transaction elements. The second vocabulary mapping module outputs multiple extracted transaction elements (such as counterparty name, transaction direction, transaction period, type of transaction instrument, transaction quantity, etc.) and transaction price.

[0266] For example, the input to the second vocabulary mapping module contains a sequence of masking information. The second vocabulary mapping module can automatically remove the masking part and placeholders from the output sequence and convert it into words represented by characters through a vocabulary list to form the final natural language sequence output, i.e., multiple transaction elements.

[0267] For example, the transaction element extraction model can identify simple spelling errors and automatically correct them, then use the corrected text for semantic analysis and output transaction elements. This demonstrates that the transaction element extraction model possesses a certain generalization ability to handle data outside the training distribution.

[0268] The hyperparameters of the transaction element extraction model in this embodiment include, but are not limited to: the vector dimension output by the word embedding module, the number of model structure layers, the number of neurons in the normalization module (including the first and second normalization modules), and the activation function type. There are multiple potential combinations of model hyperparameter settings, requiring adjustments to individual hyperparameters and combinations of hyperparameters, followed by retraining and result verification to optimize the hyperparameter settings.

[0269] like Figure 5 The diagram shown is a structural schematic of the first gating mechanism module or the second gating mechanism module provided in an embodiment of this application. Figure 5 As shown, the first gating mechanism module or the second gating mechanism module includes:

[0270] The first fully connected layer 501 has an input terminal that is either the first gating mechanism module or the second gating mechanism module.

[0271] A first activation function 502 is connected to the output of the first fully connected layer, and the output of the first activation function is the output of the first gating mechanism module or the second gating mechanism module.

[0272] For example, the first activation function is the Sigmoid activation function.

[0273] It is understandable that in the transaction element extraction model, the splicing module and the first residual connection module are connected through the first gating mechanism; the first residual connection module and the second residual connection module are connected through the second gating mechanism module.

[0274] The formula of the residual connection module without introducing the gating mechanism module (the first gating mechanism module or the second gating mechanism module) is: Y = f(X) + X, where X is the input of the gating mechanism module, and f(X) is the output after transformation by some layers (such as convolutional layers, fully connected layers, etc.). The formula of the residual connection module with the introduction of the gating mechanism module is changed to: Y = f(X) + g(X); where g(X) is a gating factor used to adjust the strength of the residual connection. The gating factor is realized by first performing a linear transformation on the input X and then applying a sigmoid activation function, that is, where Wg and bg are learnable weight and bias parameters, and σ represents the sigmoid function.

[0275] By learning the gating factor g(X), the transaction element extraction model can dynamically adjust the contribution of the residual connection according to the characteristics of the input data. When g(X) is close to 1, the contribution of the residual connection is larger; when g(X) is close to 0, the contribution of the residual connection is smaller. This dynamic adjustment capability makes the transaction element extraction model more flexible when processing different types of inputs, and can better adapt to complex patterns and changes.

[0276] Traditional residual connection is linear, while the gating mechanism introduces a nonlinear activation function (sigmoid), which increases the nonlinear expression capability of the transaction element extraction model. The improvement of nonlinear expression capability enables the transaction element extraction model to learn more complex feature representations and data relationships, thereby improving the fitting and understanding ability of complex data.

[0277] In some cases, the transaction element extraction model may rely too much on the residual connection, causing the contribution of some layers to be ignored. The gating mechanism controls the strength of the residual connection, preventing this from happening, so that each layer can play a reasonable role in the training and inference process of the transaction element extraction model, which helps to stably train deep models.

[0278] As shown in FIG. 1, a structure schematic diagram of a filtering module provided by an embodiment of the present application is shown, which includes: Figure 6 A second fully connected layer 601, an input end of the second fully connected layer being an input end of the filtering module;

[0279] A second activation function 602 connected to an output end of the second fully connected layer; an output end of the second activation function being an output end of the filtering module.

[0280] For example, the second activation function is a Swish activation function.

[0281]

[0282] ​It can be understood that the splicing module and the weighted cross-layer connection module in the transaction element extraction model are connected across layers through the filtering module. The cross-layer connection enhances the reference to the input vector, reducing the overfitting of the transaction element extraction model. At the same time, in the cross-layer connection, Swish is used as the second activation function, so that the output is not limited to the range of [0, 1], and therefore richer information can be provided between negative values and positive values. When the output is 0, it means complete neglect, and when the output is negative, it means reversing the input. It can produce a non-zero output in the negative input area. It is more flexible in response to input, and can maintain a certain flow of information in both positive and negative value areas, promoting the training of deeper networks.

[0283] The above introduces a transaction method for transaction element identification based on a large language model provided by an embodiment of the application. The device for executing the transaction method for transaction element identification based on a large language model will be introduced below.

[0284] Please refer to Figure 7 , Figure 7 The structure diagram of a transaction device for transaction element identification based on a large language model provided by an embodiment of the application is shown in FIG. 1. As shown in Figure 7 The transaction device for transaction element identification based on a large language model includes:

[0285] The first acquisition module 701 is configured to acquire a to-be-tested negotiation transaction intention text in a text dialogue window, wherein the to-be-tested negotiation transaction intention text includes a plurality of transaction elements and a transaction price.

[0286] The first generation module 702 is configured to generate transaction record information based on the to-be-tested negotiation transaction intention text, wherein the transaction record information includes the plurality of transaction elements and the transaction price.

[0287] The second acquisition module 703 is configured to acquire target transaction record information matching transaction elements in an inquiry transaction information from a plurality of the transaction record information if the inquiry transaction information including the transaction elements is received.

[0288] The second generation module 704 is configured to generate a counter-offer transaction information including the transaction price in the target transaction record information.

[0289] The first sending module 705 is configured to send the counter-offer transaction information to a first terminal held by an inquiring party.

[0290] The first receiving module 706 is configured to receive a transaction information indicating that the inquiring party determines that the transaction price in the target transaction record information is correct.

[0291] The third generation module 707 is configured to generate the confirmation transaction information if the transaction price in the transaction information is the same as the transaction price in the target transaction record information.

[0292] In an optional implementation, the first generation module includes:

[0293] The extraction unit is configured to extract the plurality of transaction elements and the transaction price from the to-be-tested negotiation transaction intention text.

[0294] The control display unit is configured to control display of the transaction elements and the transaction price.

[0295] The generation unit is configured to generate the transaction record information in response to the confirmation of no error.

[0296] In an optional implementation, the extraction unit includes:

[0297] The acquisition subunit is configured to input the to-be-tested negotiation transaction intention text into a pre-constructed transaction element extraction model, and acquire the transaction elements and the transaction price in the to-be-tested negotiation transaction intention text by using the transaction element extraction model.

[0298] In an optional implementation, the method further includes:

[0299] The third acquisition module is configured to input the simulated negotiation transaction intention text into the transaction element extraction model, and output an extraction element result by using the transaction element extraction model, the extraction element result including a plurality of transaction elements and a transaction price.

[0300] The fourth acquisition module is configured to acquire difference information between a labeled result of the simulated negotiation transaction intention text and the extraction element result, the labeled result including a plurality of transaction elements and a transaction price.

[0301] The training module is configured to train the transaction element extraction model based on the difference information.

[0302] In an optional implementation, the plurality of transaction elements include a transaction counterparty name, a transaction price, a transaction term, a transaction date, a transaction direction, and a transaction variety type, and further include:

[0303] The fifth acquisition module is configured to randomly acquire a target transaction counterparty name from a plurality of preset transaction counterparty names.

[0304] The sixth acquisition module is configured to randomly acquire an amount value within a preset amount data range.

[0305] The first determination module is configured to determine that a preset amount identifier, the amount value, an amount unit, and a first preset interference character constitute a target transaction price.

[0306] a seventh obtaining module, configured to obtain a term value randomly in a preset term data range;

[0307] a second determining module, configured to determine that the term value, a term unit and a second preset interference character constitute a target transaction date;

[0308] an eighth obtaining module, configured to obtain a pending transaction direction randomly from a plurality of preset transaction directions;

[0309] a third determining module, configured to determine that the pending transaction direction and a third preset interference character constitute a target transaction direction;

[0310] a combination module, configured to combine the target transaction counterpart name, the target transaction price, the target transaction date and the target transaction direction to obtain the simulated negotiation transaction intention text.

[0311] In an optional implementation, the transaction element extraction model comprises:

[0312] a first vocabulary mapping module, an input end of the first vocabulary mapping module being an input end of the transaction element extraction model;

[0313] a vocabulary embedding module, an input end of the vocabulary embedding module being connected with an output end of the first vocabulary mapping module, and an output end of the vocabulary embedding module being connected with an input end of a concatenation module;

[0314] a position embedding module, an input end of the position embedding module being connected with the output end of the first vocabulary mapping module, and an output end of the position embedding module being connected with the input end of the concatenation module;

[0315] the concatenation module, an output end of the concatenation module being connected with an input end of a first normalization module, an input end of a first gating mechanism module and an input end of a filtering module respectively;

[0316] the first normalization module, an output end of the first normalization module being connected with an input end of a masked multi-head self-attention module;

[0317] the masked multi-head self-attention module, an output end of the masked multi-head self-attention module being connected with an input end of a first DropOut module;

[0318] the first DropOut module, an output end of the first DropOut module being connected with an input end of a first residual connection module and an output end of the first gating mechanism module respectively;

[0319] The output end of the first residual connection module is connected with the input end of the second normalization module and the input end of the second gating mechanism module respectively.

[0320] The output end of the second normalization module is connected with the input end of the MLP feedforward network.

[0321] The output end of the MLP feedforward network is connected with the input end of the second DropOut module.

[0322] The output end of the second DropOut module is connected with the input end of the second residual connection module.

[0323] The input end of the second residual connection module is connected with the output end of the second gating mechanism module, and the output end of the second residual connection module is connected with the input end of the weighted cross-layer connection module.

[0324] The input end of the weighted cross-layer connection module is connected with the output end of the filtering module, and the pure output end of the weighted cross-layer connection module is connected with the input end of the linear layer.

[0325] The output end of the linear layer is connected with the input end of the Softmax layer.

[0326] The output end of the Softmax layer is connected with the input end of the second vocabulary mapping module.

[0327] The output end of the second vocabulary mapping module is the output end of the transaction element extraction model.

[0328] The filtering module.

[0329] The first gating mechanism module.

[0330] The second gating mechanism module.

[0331] In an optional implementation manner, the first gating mechanism module or the second gating mechanism module comprises:

[0332] A first full connection layer, the input end of the first full connection layer is the input end of the first gating mechanism module or the second gating mechanism module.

[0333] A first activation function connected with the output end of the first full connection layer, the output end of the first activation function is the output end of the first gating mechanism module or the second gating mechanism module.

[0334] In an optional implementation, the filtering module comprises:

[0335] a second full connection layer, an input end of the second full connection layer being an input end of the filtering module;

[0336] a second activation function connected to an output end of the second full connection layer, an output end of the second activation function being an output end of the filtering module.

[0337] The embodiments of the present application also provide an electronic device. Referring to FIG. 1, Figure 8 which shows a structural schematic diagram suitable for implementing the electronic device in the embodiments of the present application. The electronic device in the embodiments of the present application can include but is not limited to a fixed terminal such as a mobile phone, a notebook computer, a PDA (Personal Digital Assistant), a PAD (Tablet Personal Computer), a desktop computer, and the like. Figure 8 The electronic device shown is only an example and should not bring any limitation to the functions and use range of the embodiments of the present application.

[0338] As shown in FIG. 8, Figure 8 the electronic device can include a processing device (for example, a central processor, a graphics processor, and the like) 801, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 802 or a program loaded from a storage device 808 into a random access memory (RAM) 803. In a state where the electronic device is powered on, various programs and data required for operation of the electronic device are also stored in the RAM 803. The processing device 801, the ROM 802, and the RAM 803 are connected to each other through a bus 804. An input / output (I / O) interface 805 is also connected to the bus 804.

[0339] Generally, the following devices can be connected to the I / O interface 805: an input device 806 including, for example, a touch screen, a touch pad, a keyboard, a mouse, a camera, a microphone, an accelerometer, a gyroscope, and the like; an output device 807 including, for example, a liquid crystal display (LCD), a speaker, a vibrator, and the like; a storage device 808 including, for example, a memory card, a hard disk, and the like; and a communication device 809. The communication device 809 can allow the electronic device to communicate with other devices wirelessly or by wire to exchange data. Although Figure 8 The electronic device with various devices is shown, but it should be understood that it is not required to implement or have all the devices shown. More or fewer devices can be alternatively implemented or provided.

[0340] The embodiments of the present application also provide a computer program product comprising computer readable instructions which, when run on an electronic device, cause the electronic device to implement any of the transaction methods based on a large language model transaction element identification provided by the embodiments of the present application.

[0341] The embodiment of the present application also provides a computer readable storage medium, which carries one or more computer programs, and when the one or more computer programs are executed by an electronic device, the electronic device can implement any transaction method for transaction element identification based on a large language model provided by the embodiment of the present application.

[0342] It should be noted that the transaction method for transaction element identification based on a large language model and the related device provided by the present application can be used in the field of artificial intelligence or the field of finance. The above is only an example and does not limit the application field of the transaction method for transaction element identification based on a large language model and the related device provided by the present application.

[0343] In addition, it should be noted that the device embodiments described above are only schematic, and the units described as separate components can or can not be physically separated, and the components shown as units can or can not be physical units, that is, they can be located in one place, or distributed on multiple network units. Part or all of the modules can be selected according to actual needs to achieve the purpose of the present embodiment. In addition, the connection relationship between the modules in the device embodiment provided by the present application indicates that there is a communication connection between them, which can be realized as one or more communication buses or signal lines.

[0344] Through the description of the above embodiments, those skilled in the art can clearly understand that the present application can be realized by means of software and necessary general hardware, and of course, it can also be realized by special hardware including special integrated circuits, special CPUs, special memories, special components, etc. Generally, functions completed by computer programs can be easily realized by corresponding hardware, and the specific hardware structure for realizing the same function can also be various, such as analog circuit, digital circuit or special circuit, etc. However, for the present application, software program implementation is a better embodiment. Based on such understanding, the technical solutions of the present application can be embodied in the form of a software product, which is stored in a readable storage medium, such as a computer's floppy disk, U disk, mobile hard disk, ROM, RAM, magnetic disk or optical disk, etc., including a plurality of instructions to make a computer device (which can be a personal computer, training device, or network device, etc.) execute the method described in each embodiment of the present application.

[0345] In the above embodiments, all or part can be realized by software, hardware, firmware or any combination thereof. When realized by software, it can be realized in the form of a computer program product in whole or in part.

[0346] The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the processes or functions described in the embodiments of the present application are generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another computer-readable storage medium, for example, the computer instructions can be transmitted from one website, computer, training device or data center to another website, computer, training device or data center through wired (such as coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (such as infrared, wireless, microwave, etc.). The computer-readable storage medium can be any available medium that can be stored by the computer or a data storage device such as a training device, a data center, etc. integrated with one or more available media sets. The available media can be a magnetic medium (for example, a floppy disk, a hard disk, a magnetic tape), an optical medium (for example, a DVD), or a semiconductor medium (for example, a solid state disk (SSD)), etc.

Claims

1. A transaction method based on transaction element identification of a large language model, characterized in that, The method comprises: obtaining a to-be-tested negotiation transaction intention text in a text dialogue window, the to-be-tested negotiation transaction intention text comprising a plurality of transaction elements and a transaction price; generating transaction record information based on the to-be-tested negotiation transaction intention text, the transaction record information comprising the plurality of transaction elements and the transaction price; if an inquiry transaction information comprising transaction elements is received, obtaining target transaction record information matching the transaction elements in the inquiry transaction information from the plurality of transaction record information; generating a counter offer transaction information comprising the transaction price in the target transaction record information; sending the counter offer transaction information to a first terminal held by an inquiring party; receiving a transaction completion information representing that the inquiring party determines that the transaction price in the target transaction record information is correct; if the transaction price in the transaction completion information is the same as the transaction price in the target transaction record information, generating a transaction completion confirmation information. 2.The transaction method based on the transaction element identification of a large language model according to claim 1, characterized in that, The method of generating the transaction record information based on the to-be-tested negotiation transaction intention text comprises: extracting the plurality of transaction elements and the transaction price from the to-be-tested negotiation transaction intention text; controlling to display the transaction elements and the transaction price; in response to a correct confirmation operation, generating the transaction record information. 3.The transaction method based on large language model transaction element identification according to any one of claims 1 or 2, characterized in that, The method of extracting the plurality of transaction elements and the transaction price from the to-be-tested negotiation transaction intention text comprises: inputting the to-be-tested negotiation transaction intention text into a pre-constructed transaction element extraction model, and obtaining the transaction elements and the transaction price in the to-be-tested negotiation transaction intention text through the transaction element extraction model. 4.The transaction method based on the transaction element identification of a large language model according to claim 3, characterized in that, The method of training the transaction element extraction model comprises: inputting a simulated negotiation transaction intention text into the transaction element extraction model, and outputting an extraction element result comprising a plurality of transaction elements and a transaction price through the transaction element extraction model; obtaining a difference information between a labeled result of the simulated negotiation transaction intention text and the extraction element result, the labeled result comprising a plurality of transaction elements and a transaction price; training the transaction element extraction model based on the difference information. 5.The transaction method based on the transaction element identification of the large language model according to claim 4, characterized in that, The plurality of transaction elements comprise a transaction counterparty name, a transaction price, a transaction term, a transaction date, a transaction direction, and a transaction variety type, and the method of obtaining each simulated negotiation transaction intention text comprises: randomly obtaining a target transaction counterparty name from a plurality of preset transaction counterparty names; randomly obtaining an amount value within a preset amount data range; determining that a preset amount identifier, the amount value, an amount unit, and a first preset interference character constitute a target transaction price; randomly obtaining a term value within a preset term data range; determining that the term value, a term unit, and a second preset interference character constitute a target transaction date; randomly obtaining a pending transaction direction from a plurality of preset transaction directions; determining that the pending transaction direction and a third preset interference character constitute a target transaction direction; combining the target transaction counterparty name, the target transaction price, the target transaction date, and the target transaction direction to obtain the simulated negotiation transaction intention text.

6. The transaction method based on large language model transaction element identification according to any one of claims 4 or 5, characterized in that, The transaction element extraction model comprises: a first vocabulary mapping module, an input end of the first vocabulary mapping module being an input end of the transaction element extraction model; a vocabulary embedding module, an input end of the vocabulary embedding module being connected with an output end of the first vocabulary mapping module; an output end of the vocabulary embedding module being connected with an input end of a concatenation module; a position embedding module, an input end of the position embedding module being connected with an output end of the first vocabulary mapping module; an output end of the position embedding module being connected with an input end of the concatenation module; the concatenation module, an output end of the concatenation module being connected with an input end of a first normalization module, an input end of a first gating mechanism module and an input end of a filtering module respectively; the first normalization module, an output end of the first normalization module being connected with an input end of a masked multi-head self-attention module; the masked multi-head self-attention module, an output end of the masked multi-head self-attention module being connected with an input end of a first DropOut module; the first DropOut module, an output end of the first DropOut module being connected with an input end of a first residual connection module and an output end of the first gating mechanism module respectively; the first residual connection module, an output end of the first residual connection module being connected with an input end of a second normalization module and an input end of a second gating mechanism module respectively; the second normalization module, an output end of the second normalization module being connected with an input end of an MLP feedforward network; the MLP feedforward network, an output end of the MLP feedforward network being connected with an input end of a second DropOut module; the second DropOut module, an output end of the second DropOut module being connected with an input end of a second residual connection module; the second residual connection module, an input end of the second residual connection module being connected with an output end of the second gating mechanism module; an output end of the second residual connection module being connected with an input end of a weighted cross-layer connection module; the weighted cross-layer connection module, an input end of the weighted cross-layer connection module being connected with an output end of the filtering module; a pure output end of the weighted cross-layer connection module being connected with an input end of a linear layer; the linear layer, an output end of the linear layer being connected with an input end of a Softmax layer; the Softmax layer, an output end of the Softmax layer being connected with an input end of a second vocabulary mapping module; the second vocabulary mapping module, an output end of the second vocabulary mapping module being an output end of the transaction element extraction model; the filtering module; the first gating mechanism module; the second gating mechanism module. 7.The transaction method based on the transaction element identification of the large language model according to claim 6, characterized in that, The first gating mechanism module or the second gating mechanism module comprises: a first fully connected layer, an input end of the first fully connected layer being an input end of the first gating mechanism module or the second gating mechanism module; a first activation function connected with an output end of the first fully connected layer, an output end of the first activation function being an output end of the first gating mechanism module or the second gating mechanism module.

8. The transaction method based on large language model transaction element identification according to any one of claims 6-7, characterized in that, The filtering module comprises: A second full connection layer, an input end of the second full connection layer being an input end of the filtering module; A second activation function connected with an output end of the second full connection layer, an output end of the second activation function being an output end of the filtering module. 9.A transaction device for transaction element recognition based on a large language model, the transaction device comprising: The method comprises the following steps: A first obtaining module is configured to obtain a to-be-tested negotiation transaction intention text in a text dialogue window, the to-be-tested negotiation transaction intention text comprising a plurality of transaction elements and a transaction price; A first generating module is configured to generate transaction record information based on the to-be-tested negotiation transaction intention text, the transaction record information comprising the plurality of transaction elements and the transaction price; A second obtaining module is configured to, if an inquiry transaction information comprising transaction elements is received, obtain target transaction record information matching the transaction elements in the inquiry transaction information from the plurality of transaction record information; A second generating module is configured to generate a counter-offer transaction information comprising the transaction price in the target transaction record information; A first sending module is configured to send the counter-offer transaction information to a first terminal held by an inquiring party; A first receiving module is configured to receive a transaction completion information representing that the inquiring party determines that the transaction price in the target transaction record information is correct; A third generating module is configured to, if the transaction price in the transaction completion information is the same as the transaction price in the target transaction record information, generate a confirmation transaction completion information.

10. A computer program product, characterised in that, The computer-readable instructions, when executed on an electronic device, cause the electronic device to implement the transaction method based on the transaction element identification of the large language model according to any one of claims 1 to 8.

11. An electronic device, comprising: The electronic device comprises at least one processor and a memory connected with the processor, wherein: The memory is configured to store a computer program; The processor is configured to execute the computer program, so that the electronic device can implement the transaction method based on the transaction element identification of the large language model according to any one of claims 1 to 8.

12. A computer storage medium, characterized in that The storage medium carries one or more computer programs, which, when executed by an electronic device, can cause the electronic device to implement the transaction method based on the transaction element identification of the large language model according to any one of claims 1 to 8.

Images