A slot recognition method, device, equipment and medium

Abstract

This application provides a slot identification method, apparatus, device, and medium to solve the problem of inaccurate slot identification in the prior art. In this application embodiment, the electronic device determines the target intent corresponding to the information to be identified, and the initial slot corresponding to each keyword contained in the information to be identified, and obtains the target slot group corresponding to the target intent. Then, it adjusts the initial slots corresponding to the keywords according to the target slot group, so that the initial slots corresponding to the identified keywords all match the slots in the target slot group corresponding to the target intent. Furthermore, when adjusting the initial slots, it determines whether to adjust the initial slots based on whether there are corresponding conflicting slots. Therefore, accurate slot identification can be achieved.

Description

Technical Field

[0001] This application relates to the field of data processing technology, and in particular to a slot identification method, apparatus, device and medium. Background Technology

[0002] With the widespread application of the internet, the use of intelligent customer service and voice robots is also increasing. These technologies have solved problems in traditional industries such as recruitment difficulties, long training times, and high labor costs. When using intelligent customer service and voice robots, it is often necessary to identify the slots corresponding to keywords in the information to be identified. This information is the user's response, which is then used to control dialogue logic transitions and identify dialogue keywords.

[0003] However, in existing technologies, when identifying slots corresponding to keywords in the information to be identified, different slots may contain the same keyword. For example, the slot for "province" may contain the keyword "Shandong Province," and the slot for "range" may also contain the keyword "Shandong Province." Therefore, when the keyword "Shandong Province" appears, it may lead to inaccurate identification of the slot corresponding to "Shandong Province." For example, when the information to be identified is "Shandong Province's GDP ranks among sub-provincial cities," the slot corresponding to the keyword "Shandong Province" should be "province." If the slot corresponding to the keyword "Shandong Province" is identified as "range," subsequent processing such as controlling dialogue logic jumps and identifying dialogue keywords cannot be performed. Summary of the Invention

[0004] This application provides a slot identification method, apparatus, device, and medium to solve the problem of inaccurate slot identification in the prior art.

[0005] In a first aspect, embodiments of this application provide a slot identification method, the method comprising:

[0006] Determine the target intent corresponding to the information to be identified, and the initial slot corresponding to each keyword in the information to be identified;

[0007] Based on the pre-saved correspondence between intents and slots, determine the target slot group corresponding to the target intent;

[0008] For each initial slot, if the target slot group does not contain the initial slot and a first conflicting slot corresponding to the initial slot is pre-saved, then it is determined whether the target slot group contains the first conflicting slot. If so, the initial slot is adjusted to the first conflicting slot.

[0009] Secondly, embodiments of this application also provide a slot identification device, the device comprising:

[0010] The acquisition module is used to determine the target intent corresponding to the information to be identified, and the initial slot corresponding to each keyword in the information to be identified;

[0011] The determination module is used to determine the target slot group corresponding to the target intent based on the pre-saved correspondence between intents and slot groups;

[0012] The processing module is used to determine whether the target slot group contains the first conflicting slot for each initial slot if the target slot group does not contain the initial slot and the first conflicting slot corresponding to the initial slot is stored in advance. If so, the initial slot is adjusted to the first conflicting slot.

[0013] Thirdly, embodiments of this application also provide an electronic device, which includes at least a processor and a memory, wherein the processor is used to execute a computer program stored in the memory to implement the steps of the slot identification method as described in any of the preceding claims.

[0014] Fourthly, embodiments of this application also provide a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the slot identification method as described in any of the preceding claims.

[0015] In this embodiment, the electronic device determines the target intent corresponding to the information to be identified, and the initial slot corresponding to each keyword in the information to be identified. Based on a pre-saved correspondence between intents and slot groups, the electronic device determines the target slot group corresponding to the target intent. For each initial slot, if the target slot group does not contain the initial slot, and a first conflicting slot corresponding to the initial slot is pre-saved, the electronic device determines whether the target slot group contains the first conflicting slot. If the target slot group contains the first conflicting slot, the initial slot is adjusted to the first conflicting slot. Because in this embodiment, the electronic device determines the target intent corresponding to the information to be identified, and the initial slot corresponding to each keyword in the information to be identified, and obtains the target slot group corresponding to the target intent, and adjusts the initial slots corresponding to the keywords according to the target slot group, the initial slots corresponding to the identified keywords all match the slots in the target slot group corresponding to the target intent. Furthermore, when adjusting the initial slots, the device determines whether to adjust the initial slots based on whether there are corresponding conflicting slots, thus accurately achieving slot identification. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 This is a schematic diagram of a slot identification process provided in an embodiment of this application;

[0018] Figure 2 This is a schematic diagram illustrating a slot determination process provided in an embodiment of this application;

[0019] Figure 3 This is a schematic diagram illustrating a slot determination process provided in an embodiment of this application;

[0020] Figure 4 This is a schematic diagram illustrating a slot determination process provided in an embodiment of this application;

[0021] Figure 5 A schematic diagram illustrating the process of determining the correspondence between slots and conflicting slots, provided in an embodiment of this application;

[0022] Figure 6 This is a schematic diagram of a first model training process provided in an embodiment of this application;

[0023] Figure 7 This is a schematic diagram of a second model training process provided in an embodiment of this application;

[0024] Figure 8 This is a schematic diagram of a third model training process provided in an embodiment of this application;

[0025] Figure 9 This is a schematic diagram of a slot identification device provided in an embodiment of this application;

[0026] Figure 10 This is a schematic diagram of an electronic device structure provided in an embodiment of this application. Detailed Implementation

[0027] The present application will now be described in further detail with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments of the present application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present application.

[0028] In this embodiment of the application, the electronic device determines the target intent corresponding to the information to be identified, and the initial slot corresponding to each keyword in the information to be identified. The electronic device determines the target slot group corresponding to the target intent according to the pre-saved correspondence between intent and slot group. For each initial slot, if the target slot group does not contain the initial slot, and a first conflicting slot corresponding to the initial slot is pre-saved, it is determined whether the target slot group contains the first conflicting slot. If the target slot group contains the first conflicting slot, the initial slot is adjusted to the first conflicting slot.

[0029] To accurately identify slot positions, embodiments of this application provide a slot identification method, apparatus, device, and medium.

[0030] Figure 1 This application provides a schematic diagram of a slot identification process, which includes the following steps:

[0031] S101: Determine the target intent corresponding to the information to be identified, and the initial slot corresponding to each keyword in the information to be identified.

[0032] The slot identification method provided in this application is applied to electronic devices, such as devices with intelligent customer service applications, intelligent robots, terminals, PCs, or servers.

[0033] In this embodiment, when a user needs to interact with an electronic device or query certain information, they can input the information to be recognized through a preset page on their device. The electronic device can then obtain this information. The user's device can be an electronic device itself or a device connected to the electronic device. Alternatively, a data acquisition device can collect the user's voice information and send it to the electronic device. Upon receiving the voice information, the electronic device can identify it as the information to be recognized. The electronic device can also convert the voice information into text format and use this text format information as the information to be recognized. The data acquisition device can be an external device connected to the electronic device.

[0034] After acquiring the information to be identified, the electronic device can determine the target intent corresponding to the information and the initial slots corresponding to each keyword in the information. The electronic device can input the information to be identified into a pre-trained model and obtain the output of the pre-trained model. This output includes the target intent corresponding to the information to be identified and the initial slots corresponding to each keyword in the information. Different identifiers can be used to distinguish the target intent and the initial slots corresponding to each keyword in the output. The electronic device can pre-store the identifiers corresponding to the intent and the identifiers corresponding to the slots, and thus determine the target intent corresponding to the information to be identified and the initial slots corresponding to each keyword in the information to be identified based on the pre-stored identifiers. The target intent refers to the purpose of the information to be identified, that is, what the information to be identified wants to do. This target intent includes: checking rankings, checking metrics, etc. The initial slots are equivalent to the types corresponding to the keywords in the information to be identified, and the initial slots include: saving, time, scope, etc.

[0035] The application scenario can be a city-based question-and-answer scenario. Taking the question "What is the ranking of Qingdao's GDP in Shandong Province?" as an example, the electronic device determines that the target intent corresponding to the question is "check ranking". The initial slot corresponding to the keyword "Qingdao" in the question is "city", the initial slot corresponding to the keyword "GDP" is "indicator", and the initial slot corresponding to the keyword "Shandong Province" is "province". Taking the question "What is the GDP of Qingdao, Shandong Province?" as an example, the electronic device determines that the target intent corresponding to the question is "check indicator". The initial slot corresponding to the keyword "Qingdao" in the question is "city", the initial slot corresponding to the keyword "GDP" is "indicator", and the initial slot corresponding to the keyword "Shandong Province" is "province".

[0036] S102: Determine the target slot group corresponding to the target intent based on the pre-saved correspondence between intents and slot groups.

[0037] In this embodiment, the electronic device pre-stores a correspondence between intents and slot groups. Each slot group contains at least one slot. In the correspondence between intents and slot groups, each slot in the slot group corresponding to a certain intent is the slot required when querying that intent. Specifically, the keyword corresponding to each slot is required. For example, in an urban traffic scenario, the slot group corresponding to the intent "check ranking" is "city," "indicator," and "range," and each slot in this slot group contains "city," "indicator," and "range," respectively.

[0038] In this embodiment of the application, after the electronic device obtains the target intent corresponding to the information to be identified, it can determine each slot required when querying the target intent. Therefore, the electronic device can determine the target slot group corresponding to the target intent according to the pre-saved correspondence between intent and slot group.

[0039] Table 1 shows the correspondence between some intentions and slot groups provided in the embodiments of this application.

[0040] intention Slot group Check indicators [City], [Province], [Indicator] Check rankings [City], [Scope], [Indicators]

[0041] Table 1

[0042] In Table 1, the first column refers to the correspondence between intents and slot groups. The second column refers to the slot groups. The second row of Table 1 indicates that the slot group corresponding to the intent "Check Indicators" is "City," "Province," and "Indicators," and the slots contained in this slot group are "City," "Province," and "Indicators," respectively. The third row of Table 1 indicates that the slot group corresponding to the intent "Check Ranking" is "City," "Scope," and "Indicators," and the slots contained in this slot group are "City," "Province," and "Indicators," respectively.

[0043] In the embodiments of this application, the slot groups corresponding to other intentions may include not only "province" and "city", but also slots corresponding to smaller locations, such as the slot for "district".

[0044] S103: For each initial slot, if the target slot group does not contain the initial slot and the first conflicting slot corresponding to the initial slot is pre-saved, then determine whether the target slot group contains the first conflicting slot. If so, then adjust the initial slot to the first conflicting slot.

[0045] In practical applications, different slots may contain the same keyword, potentially leading to misidentification when determining the initial slot corresponding to that keyword. Therefore, in this embodiment, the electronic device pre-stores a correspondence between slots and conflicting slots. Any slot and its corresponding conflicting slot in this correspondence represent the same keyword. When determining the initial slot for each keyword in the information to be identified, if the initial slot for a particular keyword is incorrectly identified, it will typically be incorrectly identified as the conflicting slot corresponding to that initial slot. For example, if the correspondence includes a relationship between "province" and "scope," the accurate slot is the keyword "province." When determining the initial slot corresponding to this keyword, it is usually determined as either "province" or "scope."

[0046] In this embodiment of the application, after determining the initial slot corresponding to each keyword in the information to be identified and obtaining the target slot group corresponding to the target intent, the electronic device can determine whether the target slot group corresponding to the target intent contains the initial slot for each initial slot. If the target slot group corresponding to the target intent does not contain the initial slot, it indicates that there is an error in the determination of the initial slot. The electronic device further determines whether the pre-saved correspondence between slots and conflicting slots contains the initial slot. If the pre-saved correspondence between slots and conflicting slots contains the initial slot, the electronic device obtains the first conflicting slot corresponding to the initial slot in the correspondence between slots and conflicting slots. The first conflicting slot may be the accurate slot of the keyword corresponding to the initial slot.

[0047] To further determine whether the first conflicting slot is an accurate slot, the electronic device checks whether the determined target slot group contains the first conflicting slot. If the determined target slot group contains the first conflicting slot, it further indicates that the first conflicting slot is an accurate slot, and the electronic device adjusts the initial slot to the first conflicting slot. In this way, the electronic device can determine the accurate slot corresponding to each keyword in the information to be identified.

[0048] Taking the information to be identified as "Qingdao's Gross Domestic Product (GDP) ranks in Shandong Province" as an example, the electronic device can determine that the target intent corresponding to the information to be identified is "check ranking". It can determine that the initial slot corresponding to the keyword "Qingdao" is "city", the initial slot corresponding to the keyword "GDP" is "indicator", and the initial slot corresponding to the keyword "Shandong Province" is "province". According to the correspondence between intent and slot group, the electronic device determines that the target slot group corresponding to the target intent "check ranking" is "city", "indicator", and "scope". That is to say, the slots included in the target slot group are "city", "indicator", and "scope". And the correspondence between slots and conflicting slots includes the correspondence between "province" and "scope". Then, for the initial slot "province" that has been determined, the electronic device determines that the target slot group does not contain this slot. And the correspondence between slots and conflicting slots includes the first conflicting slot "scope" corresponding to the initial slot "province". Since the target slot group contains the first conflicting slot "scope", the electronic device can adjust the initial slot "province" to "scope".

[0049] In this embodiment, when different slots contain the same keyword, forcibly training the same keyword into two different slots during model training would often result in incorrect slot determinations. Therefore, in this embodiment, to address the situation where different slots contain the same keyword, the slots corresponding to the same keyword are trained to be one type of slot during model training. For example, the "range" slot includes keywords such as "Shandong Province" and "coastal cities," while the "province" slot includes keywords such as "Shandong Province." The "range" slot contains both keywords that are the same as those in the "province" slot and keywords that are different. Therefore, during model training, the slots corresponding to keywords like "Shandong Province" can be trained to be "province." Subsequently, each slot determined by the model is adjusted based on the target slot group corresponding to the target intent. This embodiment provides an accurate way to identify slots when different slots contain the same keyword, thereby improving the accuracy of slot determination.

[0050] In this embodiment, the electronic device determines the target intent corresponding to the information to be identified, the initial slot corresponding to each keyword in the information to be identified, and obtains the target slot group corresponding to the target intent. Then, it adjusts the initial slot corresponding to the keyword according to the target slot group, so that the initial slot corresponding to the identified keyword matches the slot in the target slot group corresponding to the target intent. Furthermore, when adjusting the initial slot, it determines whether to adjust the initial slot based on whether there is a corresponding conflicting slot. Therefore, the identification of the slot can be accurately achieved.

[0051] To accurately identify slots in the information to be identified, based on the above embodiments, the method in this application embodiment further includes:

[0052] For each initial slot, if the target slot group contains that initial slot, then the initial slot is determined to be the accurate slot.

[0053] In this embodiment of the application, after the electronic device determines the initial slot corresponding to each keyword in the information to be identified and determines the target slot group corresponding to the target intent, the electronic device can determine whether the target slot group corresponding to the target intent contains the initial slot for each initial slot. If the target slot group corresponding to the target intent contains the initial slot, it means that the initial slot is an accurate slot, and the electronic device does not need to adjust the initial slot.

[0054] Taking the initial slot corresponding to a certain keyword in the information to be identified by the electronic device as "indicator" and the target slot group corresponding to the target intent as "city", "indicator" and "range" as an example, the electronic device can determine that the target slot group corresponding to the target intent contains the initial slot "indicator", so the electronic device does not need to adjust the initial slot.

[0055] Figure 2 This application provides a schematic diagram of a slot determination process, which includes the following steps:

[0056] S201: Receive information to be identified.

[0057] S202: Determine the target intent corresponding to the information to be identified, and the initial slot corresponding to each keyword in the information to be identified.

[0058] S203: Determine the target slot group corresponding to the target intent based on the pre-saved correspondence between intents and slot groups.

[0059] S204: For each initial slot, determine whether the target slot group contains the initial slot. If yes, execute S205; otherwise, execute S206.

[0060] S205: Determine that the initial slot is the accurate slot.

[0061] S206: If a first conflicting slot corresponding to the initial slot is pre-saved, and the target slot group contains the first conflicting slot, then the initial slot is adjusted to the first conflicting slot.

[0062] In order to accurately identify the slots in the information to be identified, based on the above embodiments, the method in this application embodiment further includes:

[0063] For each initial slot, if the target slot group contains the initial slot and the initial slot appears at least twice in each initial slot corresponding to the information to be identified, then the pre-saved second conflicting slot corresponding to the initial slot is obtained, and it is determined whether the target slot group contains the second conflicting slot. If so, at least one of the initial slots is adjusted to the second conflicting slot.

[0064] In this embodiment of the application, the information to be identified may contain two identical keywords with different actual slots. For example, in the information to be identified, "What is the GDP ranking of Qingdao City in Shandong Province?", the accurate slot corresponding to the first "Shandong Province" is "province", and the accurate slot corresponding to the second "Shandong Province" is "range". When the electronic device determines the initial slot corresponding to the keyword in the information to be identified, it may determine both initial slots corresponding to "Shandong Province" as "province" or both as "range". Since the target slot group corresponding to the target intent of the information to be identified includes both "province" and "range", for each determined initial slot, if the target slot group contains the initial slot, the initial slot is determined to be the accurate slot, which may lead to identification errors.

[0065] Therefore, in this embodiment, the electronic device determines whether the target slot group corresponding to the obtained target intent contains the initial slot for each initial slot. If the target slot group contains the initial slot, it further obtains the number of times the initial slot appears in each initial slot corresponding to the information to be identified, and determines whether the number is at least twice. If the number is at least twice, it indicates that the initial slot may be abnormal. Then, according to the pre-saved correspondence between slots and conflicting slots, it determines the second conflicting slot corresponding to the initial slot, and determines whether the target slot group corresponding to the target intent contains the second conflicting slot. If the target slot group corresponding to the target intent contains the second conflicting slot, at least one of the initial slots corresponding to each initial slot corresponding to the information to be identified is adjusted to the second conflicting slot. This can be done by adjusting any one of the initial slots corresponding to each initial slot corresponding to the information to be identified to the second conflicting slot, or by adjusting the initial slot according to the order of each initial slot corresponding to the information to be identified and the order of each slot contained in the target slot group corresponding to the target intent. Specifically, the electronic device can obtain the order of the second conflicting slots in the target slot group, and adjust the slots in each initial slot that have the same order as the second conflicting slots to the second conflicting slots.

[0066] Taking the information to be identified as "where does Qingdao City's GDP rank in Shandong Province?" as an example, the determined target slot group is "province", "city", "indicator", and "scope". The initial slots corresponding to each keyword are as follows: the initial slot for the keyword "Shandong Province" is "province", the initial slot for the keyword "Qingdao City" is "city", the initial slot for the keyword "GDP" is "indicator", and the initial slot for the keyword "Shandong Province" is "province". Regarding the first initial slot corresponding to "Shandong Province", since the target slot group contains this initial slot, and this initial slot appears twice in the slots corresponding to the information to be identified, and the target slot group also contains this initial slot... The electronic device can adjust either of the two initial slots corresponding to the "province" in the information to be identified to the "range". The electronic device can also adjust the initial slot of "province" according to the order of the initial slot of "province" in each initial slot corresponding to the information to be identified and the order of the slot of "range" in the target slot group. Since the initial slot of "province" appearing for the second time in the information to be identified corresponds to the order 4 in each initial slot corresponding to the information to be identified, and the corresponding second conflicting slot "range" also corresponds to the order 4 in the target slot group, the electronic device can adjust the initial slot of "province" appearing for the second time to the "range".

[0067] In order to accurately process the identification information, based on the above embodiments, the method in this application embodiment further includes:

[0068] For each initial slot, if the target slot group does not contain that initial slot, and no conflicting slots corresponding to that initial slot are pre-saved, a preset reminder message is output; or

[0069] For each initial slot, if the target slot group does not contain the initial slot and a first conflicting slot corresponding to the initial slot is pre-saved, and the target slot group does not contain the first conflicting slot, then a preset reminder message is output.

[0070] In this embodiment, after determining the initial slot corresponding to each keyword in the information to be identified and determining the target slot group corresponding to the target intent, the electronic device can determine whether the target slot group contains the initial slot for each initial slot. If the target slot group does not contain the initial slot, the electronic device can determine whether the initial slot exists in the pre-saved correspondence between slots and conflict slots. If the initial slot does not exist in the pre-saved correspondence between slots and conflict slots, it means that the first conflict slot corresponding to the initial slot was not saved in advance, which means that the initial slot is inaccurate. In this case, subsequent control dialogue logic jumps and identification of dialogue keywords cannot be performed. This may be due to an anomaly in the pre-trained model, which makes it impossible to accurately identify the initial slot corresponding to each keyword in the information to be identified, or it may be due to the information to be identified being too chaotic. Therefore, the electronic device can obtain preset reminder information and output the reminder information.

[0071] In this embodiment, the electronic device determines whether the target slot group contains the initial slot for each initial slot. If the target slot group does not contain the initial slot, the electronic device can determine whether the initial slot exists in the pre-saved correspondence between slots and conflicting slots. If the initial slot exists in the pre-saved correspondence between slots and conflicting slots, it indicates that the first conflicting slot corresponding to the initial slot may be an accurate slot. The electronic device then determines whether the target slot group contains the first conflicting slot. If the target slot group does not contain the first conflicting slot, the initial slot cannot be adjusted to be included in the target slot group, and subsequent control dialogue logic jumps and identification of dialogue keywords cannot be performed. This may be due to an anomaly in the pre-trained model, which makes it impossible to accurately identify the initial slot corresponding to each keyword in the information to be identified, or it may be due to the information to be identified being too chaotic. Therefore, the electronic device can obtain preset reminder information and output the reminder information.

[0072] The preset reminder message can be something like "There is an error and further processing is not possible". When the reminder message is output, it can be either the text message corresponding to the reminder message is displayed or the voice message corresponding to the reminder message is played.

[0073] Figure 3 This application provides a schematic diagram of a slot determination process, which includes the following steps:

[0074] For any initial slot, if the target slot group does not contain that initial slot, the following can be used: Figure 3 The initial slot is determined in this way.

[0075] S301: Determine whether the initial slot is included in the pre-saved correspondence between slots and conflicting slots. If not, execute S302; if yes, execute S303.

[0076] S302: Output preset reminder information.

[0077] S303: Obtain the first conflicting slot that was previously saved for the initial slot, and execute S304.

[0078] S304: Determine whether the target slot group corresponding to the target intent contains the first conflicting slot. If yes, execute S305; otherwise, execute S302.

[0079] S305: Adjust any of the initial slots corresponding to the information to be identified to the first conflicting slot.

[0080] In this embodiment of the application, for each initial slot determined by the electronic device, if the initial slot is included in the target slot group corresponding to the target intent, and the initial slot appears once in each initial slot corresponding to the information to be identified, then the initial slot can be determined to be an accurate slot. If the initial slot appears at least twice in each initial slot corresponding to the information to be identified, and the initial slot is not present in the pre-saved correspondence between slots and conflicting slots, or the initial slot is present in the pre-saved correspondence between slots and conflicting slots, but the target slot group corresponding to the target intent does not include the second conflicting slot corresponding to the initial slot, then a preset reminder message can be output.

[0081] Figure 4 This application provides a schematic diagram of a slot determination process, which includes the following steps:

[0082] For any initial slot, if the target slot group contains that initial slot, the following can be used: Figure 4 The slot is determined in this way.

[0083] S401: Obtain the number of times the initial slot in the information to be identified has been identified.

[0084] S402: Determine if the number of times is at least two. If not, execute S403; if yes, execute S404.

[0085] S403: Determine that the initial slot is the accurate slot.

[0086] S404: Determine whether the initial slot is included in the pre-saved correspondence between slots and conflicting slots. If not, execute S405; if yes, execute S406.

[0087] S405: Outputs a preset reminder message.

[0088] S406: Obtain the second conflicting slot corresponding to the initial slot and execute S407.

[0089] S407: Determine whether the target slot group corresponding to the target intent contains the second conflicting slot. If yes, execute S408; otherwise, execute S405.

[0090] S408: Adjust the initial slot to the second conflicting slot.

[0091] To accurately determine the slots in the information to be identified, based on the above embodiments, in this embodiment, the pre-saved correspondence between slots and conflicting slots is determined in the following way:

[0092] For each pre-saved slot, based on the pre-saved correspondence between slots and keywords, obtain each keyword corresponding to that slot. If any keyword in each slot exists in the keywords corresponding to other slots, then determine that other slots and this slot are conflicting slots.

[0093] To accurately determine the slots in the information to be identified, the electronic device pre-stores the correspondence between slots and conflicting slots. Specifically, when determining the correspondence, the electronic device pre-stores each slot and its corresponding keyword. For each pre-stored slot, the electronic device retrieves the keyword corresponding to that slot based on the pre-stored correspondence. For each keyword, it determines whether the keyword exists in the keywords corresponding to other slots. If the keyword exists in the keywords corresponding to other slots, it means that the slot corresponding to the keyword could be either this slot or another slot, indicating that this slot and the other slots are conflicting slots. The electronic device can then determine that the other slot is the conflicting slot corresponding to this slot, and thus determine that this slot is the conflicting slot corresponding to the other slot.

[0094] In this embodiment of the application, the electronic device has a pre-determined correspondence between slots and conflicting slots. Therefore, based on the correspondence between slots and conflicting slots, the slot corresponding to each keyword determined by the model can be adjusted to accurately determine the slot corresponding to the information to be identified.

[0095] Table 2 shows the correspondence between some slots and keywords provided in the embodiments of this application.

[0096] slot Keywords moon January, February, March Province Shandong Province, Hebei Province, Liaoning Province city Beijing and Tianjin scope Shandong Province, Hebei Province, Liaoning Province, coastal cities, sub-provincial cities index GDP Year 2015, 2016, 2022

[0097] Table 2

[0098] In Table 2, the first column refers to some slots in the correspondence between slots and keywords; the second column refers to some keywords in the correspondence between slots and keywords; the second row of Table 1 refers to the keywords corresponding to the slot "month," including "January, February, March"; the third row of Table 1 refers to the keywords corresponding to the slot "province," including "Shandong Province, Hebei Province, Liaoning Province"; the fourth row of Table 1 refers to the keywords corresponding to the slot "city," including "Beijing Municipality, Tianjin Municipality"; the fifth row of Table 1 refers to the keywords corresponding to the slot "scope," including "Shandong Province, Hebei Province, Liaoning Province, coastal cities, sub-provincial cities"; the sixth row of Table 1 refers to the keywords corresponding to the slot "indicator," including "GDP"; and the seventh row of Table 1 refers to the keywords corresponding to the slot "year," including "2015, 2016, 2022."

[0099] As shown in Table 2, the keywords corresponding to the slots "year", "month", "city" and "indicator" do not exist in the keywords corresponding to other slots. However, the keyword in the slot "province" exists in the slot "range". Therefore, the electronic device can determine that the slot "province" is a conflicting slot corresponding to the slot "range" and the slot "range" is a conflicting slot corresponding to the slot "province".

[0100] Figure 5 This diagram illustrates a process for determining the correspondence between slots and conflicting slots, as provided in an embodiment of this application. The process includes the following steps:

[0101] S501: Retrieve each pre-saved slot.

[0102] S502: For each slot, determine each keyword corresponding to the slot based on the pre-saved correspondence between slots and keywords.

[0103] S503: For each keyword, if the keyword exists in each keyword corresponding to other slots, then the slot is determined to be a conflicting slot with the other slots.

[0104] To accurately determine the target intent corresponding to the information to be identified, and the initial slot corresponding to each keyword in the information to be identified, based on the above embodiments, in this embodiment, determining the target intent corresponding to the information to be identified, and the initial slot corresponding to each keyword in the information to be identified, includes:

[0105] Using a pre-trained first model, the target intent corresponding to the information to be identified, and the initial slot corresponding to each keyword in the information to be identified are obtained; or

[0106] The target intent corresponding to the information to be identified is obtained through a pre-trained second model; and the initial slot corresponding to each keyword in the information to be identified is obtained through a pre-trained third model.

[0107] In this embodiment of the application, in order to determine the target intent corresponding to the information to be identified and the initial slot corresponding to each keyword in the information to be identified, the electronic device, after acquiring the information to be identified, can input the information to be identified into a pre-trained first model, obtain the output of the first model, and determine that the output of the first model is the target intent corresponding to the information to be identified and the initial slot corresponding to each keyword in the information to be identified. The first model can be referred to as a semantic understanding model.

[0108] The first model can be a simple and powerfully optimized BERT (RoBERTa) model, which has a shorter training time and is more accurate.

[0109] In this embodiment, if only the first model is used to determine the target intent corresponding to the information to be identified and the initial slot corresponding to each keyword in the information to be identified, the target intent and the initial slot corresponding to each keyword may affect each other during the training of the first model, resulting in poor training performance. Therefore, in this embodiment, the electronic device can pre-store a trained second model and a third model. The second model and the third model respectively determine the target intent corresponding to the information to be identified and the initial slot corresponding to each keyword in the information to be identified, so that they will not affect each other during model training, thereby improving the accuracy of determining the target intent and the initial slot corresponding to the keyword. Therefore, in this embodiment, after obtaining the information to be identified, the electronic device can input the information to be identified into the pre-trained second model and obtain the output of the second model. The electronic device determines that the output of the second model is the target intent corresponding to the information to be identified. The second model can be called the intent recognition model. The electronic device can also input the information to be identified into the pre-trained third model and obtain the output of the third model. The electronic device determines that the output of the third model is the initial slot corresponding to each keyword in the information to be identified. The third model can be called the slot recognition model.

[0110] In this embodiment of the application, the electronic device can adjust the information to be identified to a preset format and then input the information to be identified into the first model, or input the information to be identified into the second model and the third model respectively.

[0111] In this embodiment, when training the first model, or training the second and third models, the operator can split the sample set into a training set, a validation set, and a test set. The first model can be trained using the training set, or the second and third models can be trained using the training set. The training effect of the first model, or the training effect of the second and third models, can be viewed using the validation set. The actual learning ability of the first model, or the actual learning ability of the second and third models, can be tested using the test set. Specifically, how to train the model using the training set, validation set, and test set is existing technology and will not be elaborated upon here.

[0112] To obtain a fully trained first model, based on the above embodiments, in this embodiment, the first model is trained in the following manner:

[0113] Obtain information about any sample in the sample set, and the first slot corresponding to each keyword contained in that sample information;

[0114] Input the sample information into the original first model and obtain the second slot corresponding to each keyword contained in the sample information;

[0115] The original first model is trained based on each first slot and the corresponding second slot.

[0116] To train the first model, this embodiment of the application stores a training sample set. This sample set includes sample information containing different keywords, and also includes various sample information with different corresponding intentions. For each sample information, the sample set also stores a first slot corresponding to each keyword contained within that sample information. In this embodiment, an entity recognition (begin-inside-outside-endsingle, BIOES) scheme can be used to label the first slot corresponding to each keyword in each sample information. Specifically, how to use the BIOES scheme to label the first slot corresponding to each keyword is existing technology and will not be elaborated here.

[0117] In this embodiment, after obtaining any sample information from the sample set and the first slot corresponding to each keyword contained in the sample information, the sample information is input into the original first model, which outputs the second slot corresponding to each keyword contained in the sample information. After the original first model determines the second slot corresponding to each keyword contained in the sample information, the electronic device trains the original first model based on the first slot corresponding to each keyword in the sample information and the second slot corresponding to each keyword in the sample information output by the original first model, thus obtaining the trained first model.

[0118] The neural first model is trained using the above method, and a trained first model is obtained when preset conditions are met. These preset conditions may include: the number of samples in the sample set that match each second slot with each first slot after training the original first model is greater than a set number; or the number of iterations for training the original first model reaching a set maximum number of iterations, etc. Specifically, this application does not impose limitations on these conditions.

[0119] Figure 6 This application provides a schematic diagram of a first model training process, which includes the following steps:

[0120] S601: Obtain information about any sample in the sample set, and the first slot corresponding to each keyword contained in the sample information.

[0121] S602: Input the sample information into the original first model and obtain the second slot corresponding to each keyword in the sample information output by the original first model.

[0122] S603: Train the original first model according to each first slot and the corresponding second slot.

[0123] S604: Obtain the first model after training is complete.

[0124] Figure 7 This application provides a schematic diagram of a second model training process, which includes the following steps:

[0125] S701: Obtain information about any sample in the sample set, and the first intent corresponding to that sample information.

[0126] S702: Input the sample information into the original second model and obtain the second intent corresponding to the sample information output by the original second model.

[0127] S703: Train the original second model according to the first intention and the second intention.

[0128] S704: Obtain the second model after training.

[0129] Figure 8 This application provides a schematic diagram of a third model training process, which includes the following steps:

[0130] S801: Obtain any sample information in the sample set, the third intent corresponding to the sample information, and the third slot corresponding to each keyword contained in the sample information.

[0131] S802: Input the sample information into the original third model to obtain the fourth intent corresponding to the sample information output by the original third model, and the fourth slot corresponding to each keyword contained in the sample information output by the original third model.

[0132] S803: Train the original third model based on the third intent and the fourth intent, each third slot and the corresponding fourth slot.

[0133] S804: Obtain the third model after training is complete.

[0134] Figure 9 This is a schematic diagram of a slot identification device provided in an embodiment of this application. The device includes:

[0135] The acquisition module 901 is used to determine the target intent corresponding to the information to be identified, and the initial slot corresponding to each keyword in the information to be identified;

[0136] The determination module 902 is used to determine the target slot group corresponding to the target intent based on the pre-saved correspondence between intents and slot groups;

[0137] The processing module 903 is used to determine whether the target slot group contains the first conflicting slot for each initial slot if the target slot group does not contain the initial slot and the first conflicting slot corresponding to the initial slot is stored in advance. If so, the initial slot is adjusted to the first conflicting slot.

[0138] Furthermore, the processing module 903 is also configured to determine that the initial slot is an accurate slot if the target slot group contains the initial slot for each initial slot.

[0139] Furthermore, the processing module 903 is also configured to, for each initial slot, if the target slot group contains the initial slot and the initial slot appears at least twice in each initial slot corresponding to the information to be identified, obtain the second conflicting slot corresponding to the pre-saved initial slot, determine whether the target slot group contains the second conflicting slot, and if so, adjust at least one of the initial slots to the second conflicting slot.

[0140] Furthermore, the processing module 903 is also configured to, for each initial slot, output a preset reminder message if the target slot group does not contain the initial slot and no conflicting slot corresponding to the initial slot is saved in advance; or, for each initial slot, if the target slot group does not contain the initial slot and a first conflicting slot corresponding to the initial slot is saved in advance, and the target slot group does not contain the first conflicting slot, output a preset reminder message.

[0141] Furthermore, the processing module 903 is also used to obtain each keyword corresponding to each pre-saved slot according to the pre-saved correspondence between slots and keywords. If any keyword in each slot exists in the keywords corresponding to other slots, then the other slots and the current slot are determined to be conflicting slots.

[0142] Furthermore, the acquisition module 901 is specifically used to acquire the target intent corresponding to the information to be identified and the initial slot corresponding to each keyword in the information to be identified through a pre-trained first model; or, to acquire the target intent corresponding to the information to be identified through a pre-trained second model; and to acquire the initial slot corresponding to each keyword in the information to be identified through a pre-trained third model.

[0143] Furthermore, the processing module 903 is also used to obtain any sample information in the sample set, and the first slot corresponding to each keyword contained in the sample information; input the sample information into the original first model, and obtain the second slot corresponding to each keyword contained in the sample information; and train the original first model according to each first slot and the corresponding second slot.

[0144] Figure 10 This application provides a schematic diagram of an electronic device structure based on an embodiment of the present application. In addition to the above embodiments, this application also provides an electronic device, such as... Figure 10 As shown, it includes: processor 1001, communication interface 1002, memory 1003 and communication bus 1004, wherein processor 1001, communication interface 1002 and memory 1003 communicate with each other through communication bus 1004.

[0145] The memory 1003 stores a computer program, which, when executed by the processor 1001, causes the processor 1001 to perform the following steps:

[0146] Determine the target intent corresponding to the information to be identified, and the initial slot corresponding to each keyword in the information to be identified;

[0147] Based on the pre-saved correspondence between intents and slot groups, determine the target slot group corresponding to the target intent;

[0148] For each initial slot, if the target slot group does not contain the initial slot and a first conflicting slot corresponding to the initial slot is pre-saved, then it is determined whether the target slot group contains the first conflicting slot. If so, the initial slot is adjusted to the first conflicting slot.

[0149] Furthermore, the processor 1001 is also configured to determine, for each initial slot, if the target slot group contains the initial slot, that the initial slot is an accurate slot.

[0150] Furthermore, the processor 1001 is also configured to, for each initial slot, if the target slot group contains the initial slot and the initial slot appears at least twice in each initial slot corresponding to the information to be identified, obtain the second conflicting slot corresponding to the pre-saved initial slot, determine whether the target slot group contains the second conflicting slot, and if so, adjust at least one of the initial slots to the second conflicting slot.

[0151] Furthermore, the processor 1001 is also configured to, for each initial slot, output a preset reminder message if the target slot group does not contain the initial slot and no conflicting slot corresponding to the initial slot is saved in advance; or, for each initial slot, if the target slot group does not contain the initial slot and a first conflicting slot corresponding to the initial slot is saved in advance, and the target slot group does not contain the first conflicting slot, output a preset reminder message.

[0152] Furthermore, the processor 1001 is also configured to, for each pre-saved slot, obtain each keyword corresponding to the slot according to the pre-saved correspondence between the slot and the keyword, and if any keyword in each of the keywords exists in the keywords corresponding to other slots, then determine that the other slot and the slot are conflicting slots.

[0153] Further, the processor 1001 is specifically used to obtain the target intent corresponding to the information to be identified and the initial slot corresponding to each keyword in the information to be identified through a first model that has been pre-trained; or, to obtain the target intent corresponding to the information to be identified through a second model that has been pre-trained; and to obtain the initial slot corresponding to each keyword in the information to be identified through a third model that has been pre-trained.

[0154] Furthermore, the processor 1001 is also configured to acquire any sample information in the sample set, and a first slot corresponding to each keyword contained in the sample information; input the sample information into the original first model to acquire a second slot corresponding to each keyword contained in the sample information; and train the original first model based on each first slot and the corresponding second slot.

[0155] The communication bus mentioned in the above server can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. This communication bus can be divided into address bus, data bus, control bus, etc. For ease of illustration, only one thick line is used to represent it in the diagram, but this does not mean that there is only one bus or one type of bus.

[0156] The communication interface is used for communication between the aforementioned electronic devices and other devices.

[0157] The memory may include random access memory (RAM) or non-volatile memory (NVM), such as at least one disk storage device. Optionally, the memory may also be at least one storage device located remotely from the aforementioned processor.

[0158] The processors mentioned above can be general-purpose processors, including central processing units, network processors (NPs), etc.; they can also be digital signal processors (DSPs), application-specific integrated circuits, field-programmable gate arrays or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc.

[0159] Based on the above embodiments, this application also provides a computer-readable storage medium storing a computer program executable by an electronic device. When the program is run on the electronic device, the electronic device performs the following steps:

[0160] The memory stores a computer program that, when executed by the processor, causes the processor to perform the following steps:

[0161] Determine the target intent corresponding to the information to be identified, and the initial slot corresponding to each keyword in the information to be identified;

[0162] Based on the pre-saved correspondence between intents and slot groups, determine the target slot group corresponding to the target intent;

[0163] For each initial slot, if the target slot group does not contain the initial slot and a first conflicting slot corresponding to the initial slot is pre-saved, then it is determined whether the target slot group contains the first conflicting slot. If so, the initial slot is adjusted to the first conflicting slot.

[0164] In one possible implementation, the method further includes:

[0165] For each initial slot, if the target slot group contains that initial slot, then the initial slot is determined to be the accurate slot.

[0166] In one possible implementation, the method further includes:

[0167] For each initial slot, if the target slot group contains the initial slot and the initial slot appears at least twice in each initial slot corresponding to the information to be identified, then the pre-saved second conflicting slot corresponding to the initial slot is obtained, and it is determined whether the target slot group contains the second conflicting slot. If so, at least one of the initial slots is adjusted to the second conflicting slot.

[0168] In one possible implementation, the method further includes:

[0169] For each initial slot, if the target slot group does not contain that initial slot, and no conflicting slots corresponding to that initial slot are pre-saved, a preset reminder message is output; or

[0170] For each initial slot, if the target slot group does not contain the initial slot and a first conflicting slot corresponding to the initial slot is pre-saved, and the target slot group does not contain the first conflicting slot, then a preset reminder message is output.

[0171] In one possible implementation, the pre-saved correspondence between slots and conflicting slots is determined in the following way:

[0172] For each pre-saved slot, based on the pre-saved correspondence between slots and keywords, obtain each keyword corresponding to that slot. If any keyword in each slot exists in the keywords corresponding to other slots, then determine that other slots and this slot are conflicting slots.

[0173] In one possible implementation, determining the target intent corresponding to the information to be identified, and the initial slot corresponding to each keyword in the information to be identified, includes:

[0174] Using a pre-trained first model, the target intent corresponding to the information to be identified, and the initial slot corresponding to each keyword in the information to be identified are obtained; or

[0175] The target intent corresponding to the information to be identified is obtained through a pre-trained second model; and the initial slot corresponding to each keyword in the information to be identified is obtained through a pre-trained third model.

[0176] In one possible implementation, the first model is trained in the following manner:

[0177] Obtain information about any sample in the sample set, and the first slot corresponding to each keyword contained in that sample information;

[0178] Input the sample information into the original first model and obtain the second slot corresponding to each keyword contained in the sample information;

[0179] The original first model is trained based on each first slot and the corresponding second slot.

[0180] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0181] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to this application. It should be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0182] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0183] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0184] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of this application. Therefore, if such modifications and variations fall within the scope of the claims of this application and their equivalents, this application also intends to include such modifications and variations.

Claims

1. A slot identification method, characterized in that, The method includes: Determine the target intent corresponding to the information to be identified, and the initial slot corresponding to each keyword in the information to be identified; Based on the pre-saved correspondence between intents and slot groups, determine the target slot group corresponding to the target intent; For each initial slot, if the target slot group does not contain the initial slot and the first conflicting slot corresponding to the initial slot is pre-saved, then it is determined whether the target slot group contains the first conflicting slot. If so, the initial slot is adjusted to the first conflicting slot. For each initial slot, if the target slot group contains that initial slot, then the initial slot is determined to be an accurate slot. For each initial slot, if the target slot group contains the initial slot and the initial slot appears at least twice in each initial slot corresponding to the information to be identified, then the pre-saved second conflicting slot corresponding to the initial slot is obtained, and it is determined whether the target slot group contains the second conflicting slot. If so, at least one of the initial slots is adjusted to the second conflicting slot. The pre-saved correspondence between slots and conflicting slots is determined in the following way: For each pre-saved slot, based on the pre-saved correspondence between slots and keywords, obtain each keyword corresponding to that slot. If any keyword in each slot exists in the keywords corresponding to other slots, then determine that other slots and this slot are conflicting slots.

2. The method according to claim 1, characterized in that, The method further includes: For each initial slot, if the target slot group does not contain that initial slot, and no conflicting slots corresponding to that initial slot are pre-saved, a preset reminder message is output; or For each initial slot, if the target slot group does not contain the initial slot and a first conflicting slot corresponding to the initial slot is pre-saved, and the target slot group does not contain the first conflicting slot, then a preset reminder message is output.

3. The method according to claim 1, characterized in that, The process of determining the target intent corresponding to the information to be identified, and the initial slot corresponding to each keyword in the information to be identified, includes: Using a pre-trained first model, the target intent corresponding to the information to be identified, and the initial slot corresponding to each keyword in the information to be identified are obtained; or The target intent corresponding to the information to be identified is obtained through a pre-trained second model; and the initial slot corresponding to each keyword in the information to be identified is obtained through a pre-trained third model.

4. The method according to claim 3, characterized in that, The first model was trained in the following way: Obtain information about any sample in the sample set, and the first slot corresponding to each keyword contained in that sample information; Input the sample information into the original first model and obtain the second slot corresponding to each keyword contained in the sample information; The original first model is trained based on each first slot and the corresponding second slot.

5. A slot identification device, characterized in that, The device includes: The acquisition module is used to determine the target intent corresponding to the information to be identified, and the initial slot corresponding to each keyword in the information to be identified; The determination module is used to determine the target slot group corresponding to the target intent based on the pre-saved correspondence between intents and slot groups; The processing module is used to determine whether the target slot group contains the first conflicting slot for each initial slot if the target slot group does not contain the initial slot and the first conflicting slot corresponding to the initial slot is stored in advance. If so, the initial slot is adjusted to the first conflicting slot. The processing module is also used to determine that the initial slot is an accurate slot if the target slot group contains the initial slot for each initial slot. The processing module is further configured to, for each initial slot, if the target slot group contains the initial slot and the initial slot appears at least twice in each initial slot corresponding to the information to be identified, obtain the second conflicting slot corresponding to the pre-saved initial slot, determine whether the target slot group contains the second conflicting slot, and if so, adjust at least one of the initial slots to the second conflicting slot. The pre-saved correspondence between slots and conflicting slots is determined in the following way: For each pre-saved slot, based on the pre-saved correspondence between slots and keywords, obtain each keyword corresponding to that slot. If any keyword in each slot exists in the keywords corresponding to other slots, then determine that other slots and this slot are conflicting slots.

6. An electronic device, characterized in that, The electronic device includes at least a processor and a memory, wherein the processor is used to execute a computer program stored in the memory to implement the steps of the slot identification method as described in any one of claims 1-4.

7. A computer-readable storage medium, characterized in that, It stores a computer program that, when executed by a processor, implements the steps of the slot identification method as described in any one of claims 1-4.

Images