Information extraction method, device, server and medium
By identifying dialogue fragments and sentences in freight conversation text, and combining intent category and text location to calculate confidence, the problem of low information extraction accuracy in existing technologies is solved, and high-accuracy freight information matching is achieved.
Patent Information
- Application Number
- CN202210231153.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2022-03-10
- Publication Date
- 2024-11-12
- Estimated Expiration
- 2042-03-10
AI Technical Summary
Existing technologies have low accuracy in information extraction in freight operations. Sequence models cannot identify the most reliable and useful information, and rule base construction is lagging behind with limited applicability to various scenarios.
By identifying dialogue fragments and statements in standard dialogue text, we can determine intent categories and parameter values, combine them with text location, calculate confidence levels, determine candidate expected parameter values, and select the target expected parameter value with the highest confidence level.
It improves the accuracy of information extraction, is applicable to various scenarios, avoids dependence on business experts, and has the characteristic of rapid iteration.
Smart Images

Figure CN114611528B_ABST
Abstract
Description
Technical Field
[0001] The embodiments of the present invention relate to the field of data processing technology, and in particular to an information extraction method, device, server and medium. Background Art
[0002] In the freight business, cargo owners and drivers often need to communicate multiple times through voice or text. If the communication information can be accurately extracted, it will be helpful to better match cargo owners and drivers, thereby improving the transaction rate and system efficiency.
[0003] Currently, information extraction is mostly implemented using sequence models or rule bases. Summary of the invention
[0004] The embodiments of the present invention provide an information extraction method, device, server and medium to solve the problem of low accuracy of information extraction in existing methods.
[0005] In a first aspect, an embodiment of the present invention provides an information extraction method, the method comprising:
[0006] Determining at least one dialogue segment included in the standard dialogue text and at least one dialogue sentence included in the at least one dialogue segment;
[0007] Determining an intention category and an intention parameter value of the at least one dialogue sentence, and determining a candidate expected parameter value corresponding to the at least one dialogue segment based on the intention category and the intention parameter value;
[0008] A target expected parameter value is determined from the candidate expected parameter values according to a text position of the at least one dialogue segment in the standard dialogue text.
[0009] In a second aspect, an embodiment of the present invention provides an information extraction device, the device comprising:
[0010] A dialogue sentence determination module, used to determine at least one dialogue segment included in the standard dialogue text, and at least one dialogue sentence included in the at least one dialogue segment;
[0011] a candidate expected parameter value determination module, configured to determine an intention category and an intention parameter value of the at least one dialogue sentence, and determine a candidate expected parameter value corresponding to the at least one dialogue segment according to the intention category and the intention parameter value;
[0012] The target expected parameter value determination module is used to determine the target expected parameter value from the candidate expected parameter values according to the text position of the at least one dialogue segment in the standard dialogue text.
[0013] In a third aspect, an embodiment of the present invention provides a server, wherein the server includes:
[0014] one or more processors;
[0015] a storage device for storing one or more programs,
[0016] When the one or more programs are executed by the one or more processors, the one or more processors implement the information extraction method as described in any one of the embodiments of the present invention.
[0017] In a fourth aspect, an embodiment of the present invention provides a computer-readable medium having a computer program stored thereon, which, when executed by a processor, implements an information extraction method as described in any one of the embodiments of the present invention.
[0018] The embodiment of the present invention achieves the effect of improving the accuracy of extracting target expected parameter values from standard dialogue texts. BRIEF DESCRIPTION OF THE DRAWINGS
[0019] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for use in the embodiments are briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without creative work.
[0020] Figure 1 is a flow chart of an information extraction method provided by Embodiment 1 of the present invention;
[0021] Figure 2 A flowchart of an information extraction method provided in Embodiment 2 of the present invention;
[0022] Figure 3 A flowchart of an information extraction method provided in Embodiment 3 of the present invention;
[0023] Figure 4 A schematic diagram of the structure of an information extraction device provided in Embodiment 4 of the present invention;
[0024] Figure 5 A schematic diagram of the structure of a server provided in Embodiment 5 of the present invention. DETAILED DESCRIPTION
[0025] The embodiments of the present invention are further described in detail below in conjunction with the accompanying drawings and embodiments. It is to be understood that the specific embodiments described herein are only used to explain the embodiments of the present invention, rather than to limit the present invention. It should also be noted that, for ease of description, only structures related to the embodiments of the present invention are shown in the accompanying drawings, rather than all structures.
[0026] In the freight business, in addition to the length and type of vehicle, there are also factors such as transportation time and transportation price that determine whether the transportation can be completed. Therefore, the shipper and the driver often need to communicate multiple times through voice or text. If the communication information can be accurately extracted, it will be helpful to better match the freight for the shipper and the driver, thereby improving the transaction rate and system efficiency.
[0027] At present, information extraction from text information is mostly achieved by using sequence models or rule bases. Among them, the method of using sequence models is: input text information into the trained sequence model, classify each character in the text information through pre-established labels, thereby obtaining the label information of each character, and then complete the extraction of useful information. The method of using rule bases is: relevant business experts establish rule generation rule bases, and extract useful information from text information according to the rules in the rule base.
[0028] However, although the sequence model can extract multiple useful information from text information, it cannot determine the most useful information with the highest confidence, so the accuracy of information extraction is low. The rule base method needs to rely on relevant business experts. Due to the variety of description forms of human language, the rules can only be applied to information extraction in a few specific scenarios, not all scenarios. In addition, the construction of the rule base is lagging and cannot meet the needs of business development.
[0029] Embodiment 1
[0030] Figure 1 1 is a flowchart of an information extraction method provided by the first embodiment of the present invention. This embodiment is applicable to the case of extracting target expected parameter values from conversation texts. The method can be executed by the information extraction device provided by the embodiment of the present invention. The information extraction device can be implemented by software and / or hardware. Figure 1 As shown, the method may include:
[0031] S101: Determine at least one dialogue segment included in a standard dialogue text, and at least one dialogue sentence included in the at least one dialogue segment.
[0032] The standard conversation text refers to the text record of the conversation process between the two parties. For example, in the field of freight, it can be the text record of the conversation process between the cargo owner and the driver. When the two parties directly communicate in text form, the text-based conversation process is directly obtained as the standard conversation text; when the two parties communicate in voice form, the voice-based conversation process is converted into text-based conversation process through voice recognition and used as the standard conversation text.
[0033] A conversation segment is a collection of conversation sentences between two parties on a conversation topic. For example, in the field of freight, if the owner and the driver have a conversation on "transportation time", then "transportation time" is the conversation topic, and the conversation sentence collection related to "transportation time" is the conversation segment corresponding to "transportation time". For another example, if the owner and the driver have a conversation on "transportation price", then "transportation price" is the conversation topic, and the conversation sentence collection related to "transportation price" is the conversation segment corresponding to "transportation price". A standard conversation text includes one or more conversation segments, and a conversation segment includes one or more conversation sentences.
[0034] In one embodiment, the conversation process in the form of speech is obtained between the two parties, and speech recognition is performed on the speech-formed conversation process through ASR (Automatic Speech Recognition) to obtain the text-formed conversation process as a standard conversation text. The standard conversation text is input into a conversation segmentation model to determine at least one conversation topic included in the standard conversation text and at least one conversation segment corresponding to each conversation topic. The conversation segment segmentation model is obtained by training the model based on manually segmented conversation segments and manually annotated conversation topics in the sample conversation text.
[0035] For example, in the standard dialogue text, the dialogue topic of sentences 10 to 20 is "transportation time", the dialogue topic of sentences 21 to 30 is "transportation model", and the dialogue topic of sentences 31 to 40 is "transportation time", then sentences 10 to 20 and sentences 31 to 40 are regarded as dialogue segments of "transportation time", and sentences 21 to 30 are regarded as dialogue segments of "transportation model".
[0036] After obtaining at least one dialogue segment included in the standard dialogue text, each dialogue segment is segmented into sentences using a regular expression, a machine learning classifier, or a decision tree, and at least one dialogue sentence included in each dialogue segment is determined. For example, a dialogue segment is "I need you to deliver at 15:00 3 days after loading. I need to combine the goods halfway, so I can only deliver at 8:00 4 days after loading. Then you can deliver at 18:00 3 days after loading." Then the dialogue sentences included in the dialogue segment are "I need you to deliver at 15:00 3 days after loading", "I need to combine the goods halfway, so I can only deliver at 8:00 4 days after loading", and "Then you can deliver at 18:00 3 days after loading."
[0037] By determining at least one dialogue segment included in the standard dialogue text and at least one dialogue sentence included in at least one dialogue segment, the effect of sentence division of the standard dialogue text is achieved, laying a data foundation for the subsequent determination of the intent category and intent parameter value of the dialogue sentence.
[0038] S102. Determine the intent category and intent parameter value of the at least one dialogue sentence, and determine a candidate expected parameter value corresponding to the at least one dialogue segment based on the intent category and the intent parameter value.
[0039] Among them, the intention category of the dialogue sentence reflects the dialogue intention of the implementer of the dialogue sentence. Taking the freight field as an example, the intention category of the dialogue sentence includes but is not limited to "asking about time" (for example, the driver asks when to ship), "proposing time" (for example, the cargo owner indicates the shipping time), "confirming time" (for example, the driver confirms the time proposed by the cargo owner), "refusing" (for example, the driver refuses because he does not have a schedule, or the driver's time is too late and the cargo owner refuses), "accepting" (for example, the driver and the cargo owner agree on the shipping time), "negotiating to advance" (the driver or the cargo owner hopes to ship earlier) and "negotiating to postpone" (the driver or the cargo owner hopes to ship later), etc. Taking the freight field as an example again, the intention category of the dialogue sentence also includes but is not limited to "asking about price", "proposing price", "confirming price", "refusing", "accepting", "negotiating price increase" and "negotiating price reduction".
[0040] The intention parameter value represents the numerical parameter involved in the dialogue sentence. For example, the intention category of the dialogue sentence "1,000 yuan for one shipment" is "propose a price", and the intention parameter value of this dialogue sentence is "1,000 yuan". For another example, the intention category of the dialogue sentence "Can it be delivered 1 hour in advance?" is "negotiate in advance", and the intention parameter value of this dialogue sentence is "1 hour".
[0041] In one embodiment, the dialogue sentences of the dialogue segment are input into a trained intent recognition model, the probability of each candidate intent category is output, and the candidate intent category with the highest probability is used as the intent category of the dialogue sentence. Among them, the type of intent recognition model includes but is not limited to (Bidirectional Encoder Representations) model. In addition, the dialogue sentences of the dialogue segment are also input into a trained parameter value extraction model, and the intent parameter values included in the dialogue sentences are output. Among them, the type of parameter value extraction model includes but is not limited to BiLSTM (Bi-directional Long Short-Term Memory) model.
[0042] The dialogue sentences in any dialogue segment that belong to the initial intent category are taken as the first target dialogue sentences, for example, "propose time" and "propose price" belong to the initial intent category. And the intention parameter value in the first target dialogue sentence is taken as the initial parameter value, for example, the initial parameter value corresponding to the first target dialogue sentence "1,000 yuan for one shipment" is "1,000 yuan".
[0043] The dialogue sentences in the dialogue segment that belong to the negotiation intention category are used as the second target dialogue sentences, such as "driver negotiates to deliver in advance", "driver negotiates to postpone", "shipper negotiates to deliver in advance", "shipper negotiates to postpone", "driver negotiates to increase the price", "driver negotiates to reduce the price", "shipper negotiates to increase the price" and "shipper negotiates to reduce the price". And the intention parameter value in the second target dialogue sentence is used as the negotiation parameter value. For example, the negotiation parameter value corresponding to the second target dialogue sentence "Can it be delivered 1 hour in advance" is "1 hour".
[0044] The initial parameter value and the negotiation parameter value are subjected to four arithmetic operations to determine the candidate expected parameter value corresponding to the conversation segment. For example, assuming that the initial parameter value of any conversation segment is "1000 yuan" and the negotiation parameter value of "driver negotiates price reduction" is "100 yuan", then the candidate expected parameter value corresponding to the conversation segment is 1000-100=900 yuan.
[0045] S103: Determine a target expected parameter value from the candidate expected parameter values according to the text position of the at least one dialogue segment in the standard dialogue text.
[0046] The text position refers to the relative position of the sentences in the dialogue segment in the standard dialogue text. For example, the standard dialogue segment includes 100 dialogue sentences, and a dialogue segment includes the 10th to 20th dialogue sentences. Then the text position of the dialogue segment in the standard dialogue text is 10 / 100 to 20 / 100 = 0.1 to 0.2.
[0047] In one embodiment, the standard deviation of each candidate expected parameter value is determined based on the candidate expected parameter value corresponding to each dialogue segment, and the text position of each dialogue segment in the standard dialogue text is determined. The standard deviation of each candidate expected parameter value and the text position of the dialogue segment to which the candidate expected parameter value belongs are input into the confidence calculation model to determine the confidence of each candidate expected parameter value, and then the candidate expected parameter value with the highest confidence is used as the target expected parameter value. In other words, the smaller the standard deviation of a candidate expected parameter value and the later the text position of the dialogue segment to which it belongs, the greater the probability that the candidate expected parameter value is the target expected parameter value.
[0048] The technical solution provided by the embodiment of the present invention determines at least one dialogue segment included in the standard dialogue text and at least one dialogue sentence included in the at least one dialogue segment, determines the intent category and intent parameter value of the at least one dialogue sentence, and determines the candidate expected parameter value corresponding to the at least one dialogue segment based on the intent category and the intent parameter value, and then determines the target expected parameter value from the candidate expected parameter values based on the text position of the at least one dialogue segment in the standard dialogue text, thereby achieving the effect of determining the target expected parameter value from the candidate expected parameter values based on the text position of the dialogue segment, avoiding the problem that the prior art can only extract multiple useful information but cannot determine the useful information with the highest confidence, and improves the accuracy of extracting the target expected parameter value; moreover, the technical solution provided by the embodiment of the present invention does not rely on relevant business experts to establish rules, and has the characteristics of fast iteration and wide scenario applicability.
[0049] Based on the above embodiment, optionally, the candidate expected parameter values include candidate expected time and / or candidate expected price, the candidate expected time includes candidate consignor expected time and candidate driver expected time, and the candidate expected price includes candidate consignor expected price and candidate driver expected price.
[0050] In the field of freight, the expected time of the candidate cargo owner indicates the time when the cargo owner expects the cargo to be shipped or delivered, and the expected time of the candidate driver indicates the time when the freight driver expects the cargo to be shipped or delivered. The expected price of the candidate cargo owner indicates the total price or unit price that the cargo owner expects to transport the cargo, and the expected price of the candidate driver indicates the total price or unit price that the freight driver expects to transport the cargo.
[0051] By setting the candidate expected parameter values to include candidate expected time and / or candidate expected price, the candidate expected time includes the candidate consignor expected time and candidate driver expected time, and the candidate expected price includes the candidate consignor expected price and the candidate driver expected price, the requirement in the freight field can be met that the expected time / expected price corresponding to the consignor and the driver can be determined respectively according to the conversation fragments between the consignor and the driver.
[0052] Embodiment 2
[0053] Figure 2 This is a flow chart of an information extraction method provided in the second embodiment of the present invention. This embodiment provides a specific implementation method for determining the target driver's expected time and the target cargo owner's expected time for the above embodiment, such as Figure 2 As shown, the method may include:
[0054] S201: Determine at least one dialogue segment included in a standard dialogue text, and at least one dialogue sentence included in at least one dialogue segment.
[0055] S202: Input the dialogue sentence and its context sentence in the corresponding dialogue segment into the intention recognition model to obtain the intention category of the dialogue sentence.
[0056] The intent recognition model is obtained by training the model based on the sample dialogue sentences, the annotated intent categories of the sample dialogue sentences, and the sample context sentences of the sample dialogue sentences. The optional intent recognition model is the Bert model.
[0057] In one embodiment, the dialogue sentence and the context sentence of the dialogue sentence in the dialogue segment to which it belongs are input into the intention recognition model, and the intention recognition model outputs the probability value of each candidate intention category, and the candidate intention category with the largest probability value is used as the intention category corresponding to the dialogue sentence. The candidate intention categories include but are not limited to the time of inquiry, the time of proposal, the time of confirmation, refusal, acceptance, driver negotiation in advance, cargo owner negotiation in advance, driver negotiation in delay, and cargo owner negotiation in delay, etc.
[0058] S203. Determine the intention parameter value of the dialogue sentence, and use the intention parameter value of the dialogue sentence with the intention category of the proposal time in the dialogue segment to which the dialogue sentence belongs as the initial time.
[0059] In one embodiment, the dialogue sentence is input into a pre-trained parameter value extraction model, and the intent parameter value included in the dialogue sentence is output, wherein the parameter value extraction model can be a BiLSTM model, the loss function is a cross entropy loss function, the optimization method is an Adam method, and 10 batches are trained on the training samples. After obtaining the intent parameter value, the intent parameter value is standardized, for example, by using the CN2AN method, the intent parameter value in Chinese numeral form is converted into the intent parameter value in Arabic numeral form, thereby achieving a unified digital format of the intent parameter value.
[0060] Determine the dialogue sentence with the intent category of "propose time" from the dialogue segment, and use the intent parameter value of the dialogue sentence as the initial time. For example, the dialogue sentence "15:00 delivery" has the intent category of "propose time", and its corresponding initial time is "15:00".
[0061] S204, the intention parameter value of the dialogue sentence with the intention category of driver negotiation for advance in the dialogue segment is used as the driver's advance time, the intention parameter value of the dialogue sentence with the intention category of driver negotiation for delay is used as the driver's delay time, the intention parameter value of the dialogue sentence with the intention category of cargo owner negotiation for advance is used as the cargo owner's advance time, and the intention parameter value of the dialogue sentence with the intention category of cargo owner negotiation for delay is used as the cargo owner's delay time.
[0062] In one embodiment, a dialogue sentence with an intention category of "driver negotiates to advance" is determined from the dialogue segment, and the intention parameter value of the dialogue sentence is used as the driver's advance time. A dialogue sentence with an intention category of "driver negotiates to postpone" is determined, and the intention parameter value of the dialogue sentence is used as the driver's postponement time. A dialogue sentence with an intention category of "consignor negotiates to advance" is determined, and the intention parameter value of the dialogue sentence is used as the consignor's advance time. A dialogue sentence with an intention category of "consignor negotiates to postpone" is determined, and the intention parameter value of the dialogue sentence is used as the consignor's postponement time.
[0063] For example, assume a conversation segment is: "I need you to deliver at 8:00 tomorrow. I need to share a ride, so the delivery will be 10 hours later tomorrow. I need you to deliver at most 5 hours later tomorrow."
[0064] In this conversation segment, the intent category of "I need you to deliver at 8:00 tomorrow" is "Propose time", and the initial time is "8:00 tomorrow". The intent category of "I need to share the ride, so the delivery will be 10 hours later tomorrow" is "Driver negotiates delay", and the driver's delay time is "10 hours". The intent category of "I need you to deliver at most 5 hours later tomorrow" is "Consignor negotiates delay", and the consignor's delay time is "5 hours".
[0065] S205. Determine the expected time of the candidate driver corresponding to the conversation segment based on the initial time, the driver's advance time, and the driver's delay time, and determine the expected time of the candidate consignor corresponding to the conversation segment based on the initial time, the consignor's advance time, and the consignor's delay time.
[0066] In one embodiment, the sum of the initial time and the driver's delayed time minus the driver's advance time is used to determine the candidate driver's expected time. Also, the sum of the initial time and the cargo owner's delayed time minus the cargo owner's advance time is used to determine the candidate cargo owner's expected time.
[0067] For example, assuming the initial time is "tomorrow 8:00", the driver's delay time is "5 hours", the driver's advance time is "1 hour", the cargo owner's delay time is "10 hours", and the cargo owner's advance time is "3 hours", then the candidate driver's expected time is tomorrow 8+5-1=12:00, and the candidate cargo owner's expected time is tomorrow 8+10-3=15:00.
[0068] S206. Determine the target driver's expected time from the candidate drivers' expected time according to the text position of at least one dialogue segment in the standard dialogue text, and determine the target consignor's expected time from the candidate consignor's expected time.
[0069] In one embodiment, the standard deviation of each candidate driver's expected time is determined based on the candidate driver's expected time in all the dialogue segments, and the target driver's expected time is determined from the candidate driver's expected time based on the standard deviation and the text position of the dialogue segment. In addition, the standard deviation of each candidate consignor's expected time is determined based on the candidate consignor's expected time in all the dialogue segments, and the target consignor's expected time is determined from the candidate consignor's expected time based on the standard deviation and the text position of the dialogue segment.
[0070] Optionally, S206 includes:
[0071] According to the expected time of the candidate drivers corresponding to at least one dialogue segment, the standard deviation of the expected time of each candidate driver is determined, and according to the text position and the standard deviation, the confidence value of the expected time of each candidate driver is determined; and the expected time of the target driver is determined from the expected times of the candidate drivers according to the confidence value.
[0072] In one embodiment, the text position of each dialogue segment in the standard dialogue text is determined. The expected time of the candidate drivers corresponding to each dialogue segment is unified into the standard time, such as Beijing time. Based on all the expected times of the candidate drivers, the standard deviation of each candidate driver's expected time is calculated to determine the standard deviation of each candidate driver's expected time. The standard deviation of each candidate driver's expected time and the text position of the corresponding dialogue segment are then input into the confidence calculation model, the confidence of each candidate driver's expected time is output, and the candidate driver's expected time with the highest confidence is used as the target driver's expected time.
[0073] By determining the standard deviation of each candidate driver's expected time based on the candidate driver's expected time corresponding to at least one dialogue segment, and determining the confidence value of each candidate driver's expected time based on the text position and the standard deviation, and determining the target driver's expected time from the candidate driver's expected time based on the confidence value, the confidence of the candidate driver's expected time is calculated based on the two factors of standard deviation and text position, thereby improving the accuracy of calculating the confidence and indirectly improving the accuracy of determining the target driver's expected time.
[0074] Optionally, S206 also includes:
[0075] According to the candidate consignee's expected time corresponding to at least one dialogue segment, the standard deviation of each candidate consignee's expected time is determined, and according to the text position and the standard deviation, the confidence value of each candidate consignee's expected time is determined; and the target consignee's expected time is determined from the candidate consignee's expected time according to the confidence value.
[0076] In one embodiment, the text position of each conversation segment in the standard conversation text is determined. The candidate consignor's expected time corresponding to each conversation segment is unified into a standard time, such as Beijing time. Based on all the candidate consignor's expected time, the standard deviation of each candidate consignor's expected time is calculated to determine the standard deviation of each candidate consignor's expected time. Then, the standard deviation of each candidate consignor's expected time and the text position of the corresponding conversation segment are input into the confidence calculation model, and the confidence of each candidate consignor's expected time is output, and the candidate consignor's expected time with the highest confidence is used as the target consignor's expected time.
[0077] By determining the standard deviation of each candidate consignor's expected time according to the candidate consignor's expected time corresponding to at least one dialogue segment, and determining the confidence value of each candidate consignor's expected time according to the text position and the standard deviation, and determining the target consignor's expected time from the candidate consignor's expected time according to the confidence value, the confidence of the candidate consignor's expected time is calculated based on the two factors of standard deviation and text position, thereby improving the accuracy of calculating the confidence and indirectly improving the accuracy of determining the target consignor's expected time.
[0078] The technical solution provided by the embodiment of the present invention inputs any dialogue sentence and the context sentence of the dialogue sentence in the corresponding dialogue segment into the intention recognition model to obtain the intention category of the dialogue sentence, thereby realizing the effect of combining the context of the dialogue sentence to recognize the intention of the dialogue sentence, thereby improving the accuracy of intention recognition;
[0079] By taking the intention parameter value of the dialogue statement with the intention category of proposing time in the dialogue segment to which the dialogue statement belongs as the initial time, the intention parameter value of the dialogue statement with the intention category of driver negotiating advance in the dialogue segment as the driver advance time, the intention parameter value of the dialogue statement with the intention category of driver negotiating delay as the driver delay time, the intention parameter value of the dialogue statement with the intention category of cargo owner negotiating advance as the cargo owner advance time, and the intention parameter value of the dialogue statement with the intention category of cargo owner negotiating delay as the cargo owner delay time, the candidate driver expected time corresponding to the dialogue segment is determined according to the initial time, the driver advance time and the driver delay time, and the candidate cargo owner expected time corresponding to the dialogue segment is determined according to the initial time, the cargo owner advance time and the cargo owner delay time, thereby achieving the effect of determining the candidate driver expected time and the candidate cargo owner expected time in the dialogue segment according to the dialogue segment between the driver and the cargo owner, laying a data foundation for the subsequent determination of the target driver expected time and the target cargo owner expected time.
[0080] Embodiment 3
[0081] Figure 3This is a flowchart of an information extraction method provided by Embodiment 3 of the present invention. This embodiment provides a specific implementation method for determining the target driver's expected price and the target cargo owner's expected price for the above embodiment, such as Figure 3 As shown, the method may include:
[0082] S301: Determine at least one dialogue segment included in a standard dialogue text, and at least one dialogue sentence included in at least one dialogue segment.
[0083] S302: Input the dialogue sentence and the context sentence of the dialogue sentence in the corresponding dialogue segment into the intention recognition model to obtain the intention category of the dialogue sentence.
[0084] In one embodiment, the dialogue sentence and the context sentence of the dialogue sentence in the dialogue segment to which it belongs are input into the intention recognition model, and the intention recognition model outputs the probability value of each candidate intention category, and the candidate intention category with the largest probability value is used as the intention category corresponding to the dialogue sentence. The candidate intention categories include but are not limited to asking for price, proposing price, confirming price, rejecting, accepting, driver negotiating price increase, cargo owner negotiating price reduction, driver negotiating price increase and cargo owner negotiating price reduction, etc.
[0085] S303: Determine the intention parameter value of the dialogue sentence, and use the intention parameter value of the dialogue sentence with the intention category of proposing a price in the dialogue segment to which the dialogue sentence belongs as the initial price.
[0086] In one embodiment, a dialogue sentence with an intent category of "propose a price" is determined from the dialogue segment, and the intent parameter value of the dialogue sentence is used as the initial price. For example, the intent category of the dialogue sentence "transportation costs 1,000 yuan" is "propose a price", and its corresponding initial price is "1,000 yuan".
[0087] S304. The intention parameter values of the dialogue sentences with the intention category of driver negotiating price increase in the dialogue segment are used as the driver's price increase price, the intention parameter values of the dialogue sentences with the intention category of driver negotiating price reduction are used as the driver's price reduction price, the intention parameter values of the dialogue sentences with the intention category of cargo owner negotiating price increase are used as the cargo owner's price increase price, and the intention parameter values of the dialogue sentences with the intention category of cargo owner negotiating price reduction are used as the cargo owner's price reduction price.
[0088] In one embodiment, a dialogue sentence with an intention category of "driver negotiates price increase" is determined from the dialogue segment, and the intention parameter value of the dialogue sentence is used as the driver's price increase. A dialogue sentence with an intention category of "driver negotiates price reduction" is determined, and the intention parameter value of the dialogue sentence is used as the driver's price reduction. A dialogue sentence with an intention category of "consignor negotiates price increase" is determined, and the intention parameter value of the dialogue sentence is used as the consignor's price increase. A dialogue sentence with an intention category of "consignor negotiates price reduction" is determined, and the intention parameter value of the dialogue sentence is used as the consignor's price reduction.
[0089] For example, suppose a certain conversation segment is: "One transportation costs 1,000 yuan. Can it be reduced by 200 yuan? The maximum reduction is 100 yuan."
[0090] In this conversation segment, the intent category of "1,000 yuan for one trip" is "offer a price", and the initial price is "1,000 yuan". The intent category of "Can it be 200 yuan cheaper" is "the owner negotiates a price reduction", and the price the owner reduces is "200 yuan". The intent category of "100 yuan cheaper at most" is "the driver negotiates a price reduction", and the price the driver reduces is "100 yuan".
[0091] S305. Determine the expected price of the candidate driver corresponding to the conversation segment based on the initial price, the driver's markup price and the driver's markdown price, and determine the expected price of the candidate consignor corresponding to the conversation segment based on the initial price, the consignor's markup price and the consignor's markdown price.
[0092] In one embodiment, the sum of the initial price and the driver's markup price minus the driver's markdown price is used to determine the candidate driver's expected price. Also, the sum of the initial price and the cargo owner's markup price minus the cargo owner's markdown price is used to determine the candidate cargo owner's expected price.
[0093] For example, assuming that the initial time is "1,000 yuan", the driver's markup price is "100 yuan", the driver's price reduction price is "300 yuan", the cargo owner's markup price is "100 yuan", and the cargo owner's advance time is "200 yuan", then the candidate driver's expected price is 1,000+100-300=800 yuan, and the candidate cargo owner's expected price is 1,000+100-200=900 yuan.
[0094] S306. Determine a target driver's expected price from the candidate drivers' expected prices and determine a target consignor's expected price from the candidate consignor's expected prices based on a text position of at least one conversation segment in the standard conversation text.
[0095] In one embodiment, the standard deviation of each candidate driver's expected price is determined based on the candidate driver's expected price in all the dialogue segments, and the target driver's expected price is determined from the candidate driver's expected price based on the standard deviation and the text position of the dialogue segment. In addition, the standard deviation of each candidate consignor's expected price is determined based on the candidate consignor's expected price in all the dialogue segments, and the target consignor's expected price is determined from the candidate consignor's expected price based on the standard deviation and the text position of the dialogue segment.
[0096] Optionally, S306 includes:
[0097] The standard deviation of each candidate driver's expected price is determined based on the candidate driver's expected price corresponding to at least one dialogue segment, and the confidence value of each candidate driver's expected price is determined based on the text position and the standard deviation; the target driver's expected price is determined from the candidate driver's expected prices based on the confidence value.
[0098] In one embodiment, the text position of each dialogue segment in the standard dialogue text is determined. The expected prices of the candidate drivers corresponding to each dialogue segment are unified into a standard price, such as RMB. Based on all the expected prices of the candidate drivers, the standard deviation of each candidate driver's expected price is calculated to determine the standard deviation of each candidate driver's expected price. The standard deviation of each candidate driver's expected price and the text position of the corresponding dialogue segment are then input into a confidence calculation model, the confidence of each candidate driver's expected price is output, and the candidate driver's expected price with the highest confidence is used as the target driver's expected price.
[0099] By determining the standard deviation of each candidate driver's expected price based on the candidate driver's expected price corresponding to at least one dialogue segment, and determining the confidence value of each candidate driver's expected price based on the text position and the standard deviation, and determining the target driver's expected price from the candidate drivers' expected prices based on the confidence value, the confidence of the candidate drivers' expected prices is calculated based on the two factors of standard deviation and text position, thereby improving the accuracy of calculating the confidence and indirectly improving the accuracy of determining the target driver's expected price.
[0100] Optionally, S306 also includes:
[0101] According to the candidate consignee's expected price corresponding to at least one dialogue segment, the standard deviation of each candidate consignee's expected price is determined, and according to the text position and the standard deviation, the confidence value of each candidate consignee's expected price is determined; and the target consignee's expected price is determined from the candidate consignee's expected prices according to the confidence value.
[0102] In one embodiment, the text position of each dialogue segment in the standard dialogue text is determined. The candidate expected prices corresponding to each dialogue segment are unified into a standard price, such as RMB. Based on all the candidate expected prices, the standard deviation of each candidate expected price is calculated to determine the standard deviation of each candidate expected price. Then, the standard deviation of each candidate expected price and the text position of the corresponding dialogue segment are input into the confidence calculation model, and the confidence of each candidate expected price is output, and the candidate expected price with the highest confidence is used as the target expected price.
[0103] By determining the standard deviation of each candidate consignor's expected price based on the candidate consignor's expected price corresponding to at least one dialogue segment, and determining the confidence value of each candidate consignor's expected price based on the text position and the standard deviation, and determining the target consignor's expected price from the candidate consignor's expected prices based on the confidence value, the confidence of the candidate consignor's expected price is calculated based on the two factors of standard deviation and text position, thereby improving the accuracy of calculating the confidence and indirectly improving the accuracy of determining the target consignor's expected price.
[0104] The technical solution provided by the embodiment of the present invention determines the candidate driver expected price corresponding to the dialogue segment according to the initial price, the driver's price increase price and the driver's price reduction price, and determines the candidate consignor expected price corresponding to the dialogue segment according to the initial price, the driver's price increase price and the driver's price reduction price, and determines the candidate consignor expected price corresponding to the dialogue segment according to the initial price, the driver's price increase price and the driver's price reduction price, and achieves the effect of determining the candidate driver expected price and the candidate consignor expected price in the dialogue segment according to the dialogue segment between the driver and the consignor, and laying a data foundation for the subsequent determination of the target driver expected price and the target consignor expected price.
[0105] On the basis of the above embodiments, this embodiment further provides a text verification method, which is executed before determining at least one dialogue segment included in the standard dialogue text. The text verification method includes the following steps A, B and C:
[0106] A. Convert the original conversation speech into the original conversation text, and determine the words to be verified included in the original conversation text.
[0107] In one implementation, the original conversation speech of the two parties is obtained, and based on the ASR technology, the original conversation speech is converted into original conversation text, and the original conversation text is divided into words to determine the words to be verified included in the original conversation text.
[0108] B. Determine the pinyin information of the word to be verified, and determine at least one candidate word associated with the pinyin information from the word library.
[0109] In one embodiment, a word library of correct words is pre-built based on a specific corpus dictionary, and an association relationship between candidate words and pinyin information is built in the word library. The pinyin information of the word to be verified is determined, and matching is performed in the word library according to the pinyin information to determine at least one candidate word associated with the pinyin information of the word to be verified.
[0110] For example, the pinyin information "wushiyidun" may be associated with candidate words such as "50 tons" and "51 tons".
[0111] C. Determine the transition probability value of the at least one candidate word based on the context words of the word to be verified in the original dialogue text, and verify the word to be verified based on the transition probability value to generate a standard dialogue text.
[0112] In one embodiment, the word to be checked and its context words in the original conversation text are input into a trained transition probability calculation model, the transition probability value of each candidate word is output, and the candidate word with the largest transition probability value is used as the target word. The word to be checked is then checked based on the target word to generate a standard conversation text.
[0113] By converting the original conversation speech into the original conversation text, and determining the words to be checked included in the original conversation text, determining the pinyin information of the words to be checked, and determining at least one candidate word associated with the pinyin information from the word library, and determining the transition probability value of at least one candidate word according to the context words of the words to be checked in the original conversation text, and checking the words to be checked according to the transition probability value, a standard conversation text is generated, thereby achieving the effect of correcting erroneous words in the original conversation text and improving the accuracy and reliability of the standard conversation text.
[0114] Optionally, step C includes the following steps:
[0115] The candidate word with the largest transfer probability value is used as the target word, and the target word is matched with the word to be verified by characters; in the case of character matching failure, the word to be verified is replaced with the target word to generate a standard dialogue text.
[0116] In one embodiment, the candidate word with the largest transfer probability value is used as the target word, and the target word is matched with the word to be checked to determine whether the characters of the target word and the word to be checked are the same. If the character matching is successful, the word to be checked is determined to be correct and no error correction is required; if the character matching fails, the word to be checked in the original dialogue text is replaced with the target word to generate a standard dialogue text.
[0117] The candidate word with the largest transfer probability value is taken as the target word, and the target word is matched with the word to be verified by character; when the character matching fails, the word to be verified is replaced with the target word to generate a standard dialogue text, thereby achieving the effect of correcting the word to be verified and ensuring the accuracy and reliability of the generated standard dialogue text.
[0118] Optionally, the original conversation text may be checked for repeated words using a regular expression. For example, if "1200" is expressed as "one thousand one hundred and two", the repeated word "one thousand" is identified by a regular expression and then deleted.
[0119] Embodiment 4
[0120] Figure 4 This is a schematic diagram of the structure of an information extraction device provided in Embodiment 4 of the present invention, which can execute an information extraction method provided in any embodiment of the present invention, and has the corresponding functional modules and beneficial effects of the execution method. Figure 4 As shown, the device may include:
[0121] A dialogue sentence determination module 41, used to determine at least one dialogue segment included in the standard dialogue text, and at least one dialogue sentence included in the at least one dialogue segment;
[0122] a candidate expected parameter value determination module 42, configured to determine an intention category and an intention parameter value of the at least one dialogue sentence, and determine a candidate expected parameter value corresponding to the at least one dialogue segment according to the intention category and the intention parameter value;
[0123] The target expected parameter value determination module 43 is used to determine a target expected parameter value from the candidate expected parameter values according to the text position of the at least one dialogue segment in the standard dialogue text.
[0124] Based on the above embodiment, the candidate expected parameter values include candidate expected time and / or candidate expected price, the candidate expected time includes candidate consignor expected time and candidate driver expected time, and the candidate expected price includes candidate consignor expected price and candidate driver expected price.
[0125] Based on the above embodiment, the candidate expected parameter value determination module 42 is specifically used for:
[0126] Inputting any of the dialogue sentences and the context sentences of the dialogue sentence in the corresponding dialogue segment into the intention recognition model to obtain the intention category of the dialogue sentence;
[0127] The intention recognition model is obtained by model training based on sample dialogue sentences, the annotated intention categories of the sample dialogue sentences, and sample context sentences of the sample dialogue sentences.
[0128] Based on the above embodiment, the candidate expected parameter value determination module 42 is further configured to:
[0129] The intention parameter value of the dialogue sentence with the intention category of the proposed time in any dialogue segment is taken as the initial time;
[0130] The intention parameter value of the dialogue sentence with the intention category of driver negotiating to advance in the dialogue segment is used as the driver's advance time, the intention parameter value of the dialogue sentence with the intention category of driver negotiating to postpone is used as the driver's postpone time, the intention parameter value of the dialogue sentence with the intention category of cargo owner negotiating to advance is used as the cargo owner's advance time, and the intention parameter value of the dialogue sentence with the intention category of cargo owner negotiating to postpone is used as the cargo owner's postpone time;
[0131] The candidate driver's expected time corresponding to the conversation segment is determined based on the initial time, the driver's advance time and the driver's delayed time, and the candidate consignor's expected time corresponding to the conversation segment is determined based on the initial time, the consignor's advance time and the consignor's delayed time.
[0132] Based on the above embodiment, the candidate expected parameter value determination module 42 is further configured to:
[0133] The intention parameter value of the dialogue sentence with the intention category of proposing a price in any dialogue segment is used as the initial price;
[0134] The intention parameter value of the dialogue sentence with the intention category of driver negotiating price increase in the dialogue segment is used as the driver price increase price, the intention parameter value of the dialogue sentence with the intention category of driver negotiating price reduction is used as the driver price reduction price, the intention parameter value of the dialogue sentence with the intention category of cargo owner negotiating price increase is used as the cargo owner price increase price, and the intention parameter value of the dialogue sentence with the intention category of cargo owner negotiating price reduction is used as the cargo owner price reduction price;
[0135] The candidate driver's expected price corresponding to the conversation segment is determined based on the initial price, the driver's markup price and the driver's markdown price, and the candidate consignor's expected price corresponding to the conversation segment is determined based on the initial price, the consignor's markup price and the consignor's markdown price.
[0136] Based on the above embodiment, the target expected parameter value determination module 43 is specifically used for:
[0137] Determining a standard deviation of each candidate expected parameter value according to the candidate expected parameter value corresponding to the at least one conversation segment, and determining a confidence value of each candidate expected parameter value according to the text position and the standard deviation;
[0138] A target expected parameter value is determined from the candidate expected parameter values according to the confidence value.
[0139] Based on the above embodiment, the device further includes a word verification module, which is specifically used to:
[0140] Converting the original conversation speech into the original conversation text, and determining the words to be verified included in the original conversation text;
[0141] Determine the pinyin information of the word to be verified, and determine at least one candidate word associated with the pinyin information from the word library;
[0142] The transition probability value of the at least one candidate word is determined according to the context words of the word to be verified in the original dialogue text, and the word to be verified is verified according to the transition probability value to generate a standard dialogue text.
[0143] Based on the above embodiment, the word verification module is further used for:
[0144] Taking the candidate word with the largest transfer probability value as the target word, and performing character matching between the target word and the word to be checked;
[0145] In the event that character matching fails, the word to be verified is replaced with the target word to generate a standard dialogue text.
[0146] An information extraction device provided in an embodiment of the present invention can execute an information extraction method provided in any embodiment of the present invention, and has functional modules and beneficial effects corresponding to the execution method. For technical details not described in detail in this embodiment, please refer to an information extraction method provided in any embodiment of the present invention.
[0147] Embodiment 5
[0148] Figure 5A schematic diagram of the structure of a server provided in Embodiment 5 of the present invention. Figure 5 A block diagram of an exemplary server 500 suitable for implementing embodiments of the present invention is shown. Figure 5 The server 500 shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present invention.
[0149] like Figure 5 As shown, the server 500 is in the form of a general computing server. The components of the server 500 may include, but are not limited to: one or more processors or processing units 501, a system memory 502, and a bus 503 connecting different system components (including the system memory 502 and the processing unit 501).
[0150] Bus 503 represents one or more of several types of bus structures, including a memory bus or memory controller, a peripheral bus, an accelerated graphics port, a processor or a local bus using any of a variety of bus structures. For example, these architectures include but are not limited to Industry Standard Architecture (ISA) bus, Micro Channel Architecture (MAC) bus, Enhanced ISA bus, Video Electronics Standards Association (VESA) local bus and Peripheral Component Interconnect (PCI) bus.
[0151] The server 500 typically includes a variety of computer system readable media. These media can be any available media that can be accessed by the server 500, including volatile and non-volatile media, removable and non-removable media.
[0152] System memory 502 may include computer system readable media in the form of volatile memory, such as random access memory (RAM) 504 and / or cache memory 505. Server 500 may further include other removable / non-removable, volatile / non-volatile computer system storage media. By way of example only, storage system 506 may be used to read and write non-removable, non-volatile magnetic media ( Figure 5 not shown, usually called a "hard drive"). Although Figure 5 Not shown in the figure, a disk drive for reading and writing to a removable non-volatile disk (e.g., a "floppy disk"), and an optical disk drive for reading and writing to a removable non-volatile optical disk (e.g., a CD-ROM, a DVD-ROM, or other optical media) may be provided. In these cases, each drive may be connected to the bus 503 via one or more data medium interfaces. The memory 502 may include at least one program product having a set (e.g., at least one) of program modules that are configured to perform the functions of the embodiments of the present invention.
[0153] A program / utility 508 having a set (at least one) of program modules 507 may be stored, for example, in the memory 502, such program modules 507 including but not limited to an operating system, one or more application programs, other program modules, and program data, each of which or some combination may include an implementation of a network environment. The program modules 507 generally perform the functions and / or methods of the embodiments described herein.
[0154] The server 500 may also communicate with one or more external servers 509 (e.g., keyboards, pointing servers, displays 510, etc.), one or more servers that enable users to interact with the server 500, and / or any server that enables the server 500 to communicate with one or more other computing servers (e.g., network cards, modems, etc.). Such communication may be performed via an input / output (I / O) interface 511. In addition, the server 500 may also communicate with one or more networks (e.g., local area networks (LANs), wide area networks (WANs), and / or public networks, such as the Internet) via a network adapter 512. As shown, the network adapter 512 communicates with other modules of the server 500 via a bus 503. It should be understood that, although not shown in the figure, other hardware and / or software modules may be used in conjunction with the server 500, including but not limited to: microcode, server drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data backup storage systems.
[0155] The processing unit 501 executes various functional applications and data processing by running the program stored in the system memory 502, for example, implementing the information extraction method provided in the embodiment of the present invention, including:
[0156] Determining at least one dialogue segment included in the standard dialogue text and at least one dialogue sentence included in the at least one dialogue segment;
[0157] Determining an intention category and an intention parameter value of the at least one dialogue sentence, and determining a candidate expected parameter value corresponding to the at least one dialogue segment based on the intention category and the intention parameter value;
[0158] A target expected parameter value is determined from the candidate expected parameter values according to a text position of the at least one dialogue segment in the standard dialogue text.
[0159] Embodiment 6
[0160] Embodiment 6 of the present invention further provides a computer-readable storage medium, wherein the computer-executable instructions are used to execute an information extraction method when executed by a computer processor, the method comprising:
[0161] Determining at least one dialogue segment included in the standard dialogue text and at least one dialogue sentence included in the at least one dialogue segment;
[0162] Determining an intention category and an intention parameter value of the at least one dialogue sentence, and determining a candidate expected parameter value corresponding to the at least one dialogue segment based on the intention category and the intention parameter value;
[0163] A target expected parameter value is determined from the candidate expected parameter values according to a text position of the at least one dialogue segment in the standard dialogue text.
[0164] Of course, a storage medium containing computer executable instructions provided by an embodiment of the present invention, whose computer executable instructions are not limited to the method operations described above, can also perform related operations in an information extraction method provided by any embodiment of the present invention. The computer-readable storage medium of an embodiment of the present invention can adopt any combination of one or more computer-readable media. The computer-readable medium can be a computer-readable signal medium or a computer-readable storage medium. The computer-readable storage medium can be, for example, - but not limited to - an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination of the above. More specific examples (non-exhaustive list) of computer-readable storage media include: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In this document, a computer-readable storage medium can be any tangible medium containing or storing a program, which can be used by an instruction execution system, device or device or used in combination with it.
[0165] Computer-readable signal media may include data signals propagated in baseband or as part of a carrier wave, which carry computer-readable program code. Such propagated data signals may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. Computer-readable signal media may also be any computer-readable medium other than a computer-readable storage medium, which may send, propagate, or transmit a program for use by or in conjunction with an instruction execution system, apparatus, or device.
[0166] Program code embodied on a computer readable medium may be transmitted using any appropriate medium, including but not limited to wireless, wireline, optical fiber cable, RF, etc., or any suitable combination of the foregoing.
[0167] Computer program code for performing the operations of the present invention may be written in one or more programming languages or a combination thereof, including object-oriented programming languages such as Java, Smalltalk, C++, and conventional procedural programming languages such as "C" or similar programming languages. The program code may be executed entirely on the user's computer, partially on the user's computer, as a separate software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., via the Internet using an Internet service provider).
[0168] Note that the above are only preferred embodiments of the present invention and the technical principles used. Those skilled in the art will understand that the present invention is not limited to the specific embodiments described herein, and that various obvious changes, readjustments and substitutions can be made by those skilled in the art without departing from the scope of protection of the present invention. Therefore, although the present invention has been described in more detail through the above embodiments, the present invention is not limited to the above embodiments, and may include more other equivalent embodiments without departing from the concept of the present invention, and the scope of the present invention is determined by the scope of the appended claims.
Claims
1. An information extraction method, characterized in that: The method comprises: Determining at least one dialogue segment included in the standard dialogue text and at least one dialogue sentence included in the at least one dialogue segment; Determining the intent category and the intent parameter value of the at least one dialogue sentence, and determining, based on the intent category and the intent parameter value, candidate expected parameter values corresponding to the at least one dialogue segment; wherein the candidate expected parameter values include candidate expected time and / or candidate expected price; determining a target expected parameter value from the candidate expected parameter values according to a text position of the at least one dialogue segment in the standard dialogue text; The step of determining the target expected parameter value from the candidate expected parameter values according to the text position of the at least one dialogue segment in the standard dialogue text comprises: Determining a standard deviation of each candidate expected parameter value according to the candidate expected parameter value corresponding to the at least one conversation segment, and determining a confidence value of each candidate expected parameter value according to the text position and the standard deviation; determining a target expected parameter value from the candidate expected parameter values according to the confidence value; The determining, according to the intent category and the intent parameter value, a candidate expected parameter value corresponding to the at least one conversation segment includes: The intention parameter value of the dialogue sentence with the intention category of the proposed time in any dialogue segment is taken as the initial time; The intention parameter value of the dialogue sentence with the intention category of driver negotiating to advance in the dialogue segment is used as the driver's advance time, the intention parameter value of the dialogue sentence with the intention category of driver negotiating to postpone is used as the driver's postpone time, the intention parameter value of the dialogue sentence with the intention category of cargo owner negotiating to advance is used as the cargo owner's advance time, and the intention parameter value of the dialogue sentence with the intention category of cargo owner negotiating to postpone is used as the cargo owner's postpone time; Determine the candidate driver's expected time corresponding to the conversation segment according to the initial time, the driver's advance time, and the driver's delayed time, and determine the candidate consignor's expected time corresponding to the conversation segment according to the initial time, the consignor's advance time, and the consignor's delayed time; Alternatively, determining the candidate expected parameter value corresponding to the at least one conversation segment according to the intention category and the intention parameter value includes: The intention parameter value of the dialogue sentence with the intention category of proposing a price in any dialogue segment is used as the initial price; The intention parameter value of the dialogue sentence with the intention category of driver negotiating price increase in the dialogue segment is used as the driver price increase price, the intention parameter value of the dialogue sentence with the intention category of driver negotiating price reduction is used as the driver price reduction price, the intention parameter value of the dialogue sentence with the intention category of cargo owner negotiating price increase is used as the cargo owner price increase price, and the intention parameter value of the dialogue sentence with the intention category of cargo owner negotiating price reduction is used as the cargo owner price reduction price; The candidate driver's expected price corresponding to the conversation segment is determined based on the initial price, the driver's markup price and the driver's markdown price, and the candidate consignor's expected price corresponding to the conversation segment is determined based on the initial price, the consignor's markup price and the consignor's markdown price.
2. The method according to claim 1, characterized in that The determining of the intent category of the at least one dialogue sentence comprises: Inputting any of the dialogue sentences and the context sentences of the dialogue sentence in the corresponding dialogue segment into the intention recognition model to obtain the intention category of the dialogue sentence; The intention recognition model is obtained by model training based on sample dialogue sentences, the annotated intention categories of the sample dialogue sentences, and sample context sentences of the sample dialogue sentences.
3. The method according to claim 1, before determining at least one dialogue segment included in the standard dialogue text, further comprising: Converting the original conversation speech into the original conversation text, and determining the words to be verified included in the original conversation text; Determine the pinyin information of the word to be verified, and determine at least one candidate word associated with the pinyin information from the word library; The transition probability value of the at least one candidate word is determined according to the context words of the word to be verified in the original dialogue text, and the word to be verified is verified according to the transition probability value to generate a standard dialogue text.
4. The method according to claim 3, characterized in that: The step of verifying the words to be verified according to the transition probability values to generate a standard dialogue text includes: Taking the candidate word with the largest transfer probability value as the target word, and performing character matching between the target word and the word to be checked; In the event that character matching fails, the word to be verified is replaced with the target word to generate a standard dialogue text.
5. An information extraction device, characterized in that: The device comprises: A dialogue sentence determination module, used to determine at least one dialogue segment included in the standard dialogue text, and at least one dialogue sentence included in the at least one dialogue segment; a candidate expected parameter value determination module, configured to determine the intent category and the intent parameter value of the at least one dialogue sentence, and determine the candidate expected parameter value corresponding to the at least one dialogue segment according to the intent category and the intent parameter value; wherein the candidate expected parameter value includes a candidate expected time and / or a candidate expected price; a target expected parameter value determination module, configured to determine a target expected parameter value from the candidate expected parameter values according to a text position of the at least one dialogue segment in the standard dialogue text; Wherein, the target expected parameter value determination module is specifically used for: Determining a standard deviation of each candidate expected parameter value according to the candidate expected parameter value corresponding to the at least one conversation segment, and determining a confidence value of each candidate expected parameter value according to the text position and the standard deviation; determining a target expected parameter value from the candidate expected parameter values according to the confidence value; The candidate expected parameter value determination module is specifically used to: The intention parameter value of the dialogue sentence with the intention category of the proposed time in any dialogue segment is taken as the initial time; The intention parameter value of the dialogue sentence with the intention category of driver negotiating to advance in the dialogue segment is used as the driver's advance time, the intention parameter value of the dialogue sentence with the intention category of driver negotiating to postpone is used as the driver's postpone time, the intention parameter value of the dialogue sentence with the intention category of cargo owner negotiating to advance is used as the cargo owner's advance time, and the intention parameter value of the dialogue sentence with the intention category of cargo owner negotiating to postpone is used as the cargo owner's postpone time; Determine the candidate driver's expected time corresponding to the conversation segment according to the initial time, the driver's advance time, and the driver's delayed time, and determine the candidate consignor's expected time corresponding to the conversation segment according to the initial time, the consignor's advance time, and the consignor's delayed time; Alternatively, the candidate expected parameter value determination module is specifically used to: The intention parameter value of the dialogue sentence with the intention category of proposing a price in any dialogue segment is used as the initial price; The intention parameter value of the dialogue sentence with the intention category of driver negotiating price increase in the dialogue segment is used as the driver price increase price, the intention parameter value of the dialogue sentence with the intention category of driver negotiating price reduction is used as the driver price reduction price, the intention parameter value of the dialogue sentence with the intention category of cargo owner negotiating price increase is used as the cargo owner price increase price, and the intention parameter value of the dialogue sentence with the intention category of cargo owner negotiating price reduction is used as the cargo owner price reduction price; The candidate driver's expected price corresponding to the conversation segment is determined based on the initial price, the driver's markup price and the driver's markdown price, and the candidate consignor's expected price corresponding to the conversation segment is determined based on the initial price, the consignor's markup price and the consignor's markdown price.
6. A server, characterized in that: The server also includes: one or more processors; a storage device for storing one or more programs, When the one or more programs are executed by the one or more processors, the one or more processors implement the information extraction method as described in any one of claims 1 to 4.
7. A computer readable medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the information extraction method as described in any one of claims 1 to 4 is implemented.
