Multi-time entity analysis method in semantic recognition

Abstract

The invention discloses a multi-time entity analysis method in semantic recognition, which comprises the following steps of: extracting date and time point information in an original text through entity recognition, combining time entities into one or more time nouns by utilizing understanding on the original text, and if a plurality of time nouns exist, judging whether the time nouns exist or not; and linking the related corresponding entities by using the dependency relationship of the time entities in the time nouns, constructing new time nouns, normalizing the new time nouns into absolute time, and finally generating accurate time points which can be used for query. According to the method and the system, a plurality of accurate time points are analyzed by accurately combining date and time point entities, so that the system can more accurately complete an instruction of a user, and misrecognition is reduced; the user experience of the user in the aspect of time can be improved, and interaction is more user-friendly.

Description

Technical Field

[0001] The present invention relates to the technical field of natural language semantic recognition, and in particular to a multi-time entity parsing method in semantic recognition. Background Art

[0002] In the field of natural language processing (NLP), the recognition of time entities is one of the key technologies, which is mainly divided into two categories: rule recognition and algorithm recognition. The rule recognition method relies on predefined rules to recognize time expressions in a specific format. Although this method can accurately identify time entities that meet the rules, it lacks generalization ability and requires writing a large number of complex rules for various time expressions, which is not only time-consuming but also difficult to adapt to diverse time expressions.

[0003] In contrast, algorithmic recognition methods use machine learning technology to train models using a large amount of labeled data, which can identify multiple time entities in text and have good generalization. However, these algorithms usually only provide recognition results of time entities and lack the ability to judge the association and logical relationship between time entities. Summary of the invention

[0004] The purpose of the present invention is to provide a method for parsing multiple time entities in semantic recognition in order to solve the above problems. The present invention can automatically infer the specific time period or time point corresponding to the identified time entity based on the original text content and preset logical rules. This method not only improves the accuracy and consistency of time information, but also provides clearer time data for subsequent applications, enhancing the practicality and reliability of time data processing.

[0005] The present invention achieves the above-mentioned purpose through the following technical solutions:

[0006] A method for multi-temporal entity resolution in semantic recognition, comprising the following steps:

[0007] Step 1: perform index association and intermediate text judgment on entities to obtain temporal entity recognition results;

[0008] Step 2, constructing a time noun, forming a compound time expression by combining the date and time point;

[0009] Step 3: Rules and logic deduction to resolve relative time into absolute time;

[0010] Step 4, processing the time interval formed by the composite time expression when necessary;

[0011] Step 5: Output the parsing results.

[0012] A further solution is that in step 1, using the trained model, entities related to dates and time points are identified from the text; the start and end index positions of each time entity in the text are recorded, where index 0 represents the first character in the text.

[0013] A further solution is that in step 2, according to the index positions of the time entities, entities with the same start and end indexes are combined into time nouns; if there are two time entities with a difference of 1 between the end index and the start index, it is judged whether the position of the original text located at the end index + 1 is "到" or "至", and if so, these two time entities are combined into time noun 1 and time noun 2 respectively according to the rules, and these two entities are recorded as consecutive time periods.

[0014] A further solution is that in step 3, logical reasoning rules are constructed to identify multiple time periods and non-time period time entities, and they are extended in sequence to identify multiple time periods and non-time period time entities.

[0015] A further solution is that in step 4, logical reasoning rules are constructed to analyze the mutual influence of each date and time point entity in multiple time nouns to determine how they are combined into a complete time period expression.

[0016] A further solution is that in step 4, each time noun is converted into an absolute time, while considering the relative time influence in the previous and subsequent time nouns. If a certain time entity has an impact on the previous or subsequent time noun, then this entity needs to be aligned to the corresponding time noun.

[0017] The beneficial effects of the present invention are as follows:

[0018] A method for parsing multiple time entities in semantic recognition of the present invention extracts date and time point information from the original text through entity recognition, combines time entities into one or more time nouns by understanding the original text. If there are multiple time nouns, the dependency relationship of time entities in the time nouns is used to link relevant corresponding entities to construct new time nouns, and they are normalized into absolute time, and finally an accurate time point for query can be generated. The present invention can parse out multiple accurate time points by accurately combining date and time point entities, enabling the system to more accurately complete user instructions and reduce misrecognition; it can improve the user experience in terms of time for users, making the interaction more user-friendly. Brief Description of the Drawings

[0019] To more clearly illustrate the technical solutions in the embodiments of the present invention, the following will briefly introduce the drawings required in the embodiments or the description of the prior art. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0020] Figure 1 This is the flowchart of the present invention. Specific embodiments

[0021] To make the objectives, technical solutions, and advantages of the present invention clearer, the following will describe the technical solutions of the present invention in detail. Obviously, the described embodiments are only some embodiments of the present invention, rather than all embodiments. Based on the embodiments of the present invention, all other implementation manners obtained by those of ordinary skill in the art without creative efforts fall within the scope protected by the present invention.

[0022] In any embodiment, as Figure 1 shown, a method for multi-time entity parsing in semantic recognition of the present invention includes:

[0023] Train a specific model to identify entities related to dates and time points in the text. Obtain all the mentioned date and time point entities in the text and the start and end indices of the entities in the text (the positions in the text, by default, the index of the first character in the text is 0).

[0024] According to the start and end indices of each entity, combine two entities with the same end index and start index together in sequence to form a time noun. If there are two time entities with a difference of 1 between the end index and the start index, determine whether the position of the end index + 1 in the original text is "to" or "until". If so, combine these two time entities into time noun 1 and time noun 2 respectively according to the rules, and record these two entities as consecutive time periods. This rule can be extended. As long as it conforms to the rule, multiple time periods and non-time period time entities can be identified in sequence.

[0025] Construct logical reasoning rules to judge how different date and time point entities are related to each other, and identify which entities should be combined together to form a complete time period expression.

[0026] Construct logical reasoning rules to judge whether a certain date or time point entity among multiple time nouns will have an impact on the previous or subsequent time noun. Because each time noun needs to be converted into an absolute time, this conversion needs to consider the relative time impact in each date and time point entity in the previous and subsequent time nouns. If there is an impact, the entity causing the impact needs to be aligned to the previous or subsequent time noun.

[0027] In a specific embodiment, as Figure 1 shown, a multi-time entity parsing method in semantic recognition of the present invention includes the following steps:

[0028] Problem: The system received a user instruction: "Please help me query the schedule from 12:00 noon to 4:00 pm the day after tomorrow and remind me at 12:00 on that day. At this time, entity 1 recognized by the entity recognition model has been received: {"entity type": "date", "entity value": "the day after tomorrow", "start index": "7", "end index": "9"}; entity 2: {"entity type": "time point", "entity value": "12:00 noon", "start index": "9", "end index": "14"}; entity 3: {"entity type": "time point", "entity value": "4:00 pm", "start index": "15", "end index": "19"}; entity 4: {"entity type": "date", "entity value": "that day", "start index": "27", "end index": "29"}; entity 5: {"entity type": "time point", "entity value": "12:00", "start index": "29", "end index": "32"}. It is necessary to parse the accurate absolute time for which the user wants to query the schedule and the reminder time point.

[0029] 1. First, according to the entity index, combine the entities with the same end index and start index together to construct time noun 1: 12:00 noon the day after tomorrow, time noun 2: 4:00 pm, and time noun 3: 12:00 on that day.

[0030] 2. The difference between the end index of the last entity in time noun 1 and the start index of the first entity in time noun 2 is 1. Locate the text corresponding to this difference in the user instruction, which is "to". It can be determined that time noun 1 and time noun 2 are the start time and end time of a continuous period. Since there is a date entity in time noun 1 and no date entity in time noun 2, it is judged that time noun 2 is the relative time of time noun 1. According to the time conversion rule, the date entity in time noun 1 is supplemented to time noun 2, and time noun 2 is regenerated as: "4:00 pm the day after tomorrow".

[0031] 3. According to the index position, it can be judged that time noun 3 has no association with time nouns 1 and 2. The date entity of time noun 3 is "that day". According to the entity association information, it can be judged that "that day" can be deduced to the date entity of the previous time noun, and time noun 3 is regenerated as: "12:00 the day after tomorrow".

[0032] 4. Determine that all time nouns have been converted to absolute time, and directly perform time recognition. Assuming that today is 2024-9-1, parse the time period and time point: "2024-9-3 12:00:00-2024-9-9 16:00:00"; "2024-9-3 12:00:00". Subsequent functions can use the above time to issue precise instructions to achieve user goals.

[0033] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can easily think of changes or replacements within the technical scope disclosed by the present invention, which should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims. It should also be noted that the various specific technical features described in the above specific embodiments can be combined in any suitable manner without contradiction. In order to avoid unnecessary repetition, the present invention will not further explain various possible combinations. In addition, the various different embodiments of the present invention can also be combined arbitrarily, as long as they do not violate the concept of the present invention, they should also be regarded as the contents disclosed by the present invention.

Claims

1. A method for multi-temporal entity resolution in semantic recognition, characterized in that: Including the following steps: Step 1, perform index association and intermediate text judgment on the entity to obtain the time entity recognition result; Step 2, construct time nouns, and form a composite time expression by combining the date and time point; Step 3, rule and logical derivation, resolve relative time into absolute time; Step 4, process the time interval formed by the composite time expression when necessary; Step 5, output the parsing result.

2. A method for multi-temporal entity resolution in semantic recognition as claimed in claim 1, characterized in that: In the said Step 1, use the trained model to identify the entities related to the date and time point from the text; record the start and end index positions of each time entity in the text, where index 0 represents the first character in the text.

3. The method for multi-temporal entity resolution in semantic recognition according to claim 1, characterized in that: In the said Step 2, according to the index positions of the time entities, combine the entities with the same start and end indexes into time nouns; if there are two time entities with a difference of 1 between the end index and the start index, judge whether the position of the original text located at the end index + 1 is "to" or "until", if so, combine these two time entities into time noun 1 and time noun 2 respectively according to the rules, and record these two entities as a continuous time period.

4. The method for multi-temporal entity resolution in semantic recognition according to claim 1, characterized in that: In the said Step 3, construct logical reasoning rules, identify multiple time periods and non-time period time entities, and expand them in sequence to identify multiple time periods and non-time period time entities.

5. The method for multi-temporal entity resolution in semantic recognition according to claim 1, characterized in that: In the said Step 4, construct logical reasoning rules, analyze the mutual influence of each date and time point entity in multiple time nouns to determine how they are combined into a complete time period expression.

6. A method for multi-temporal entity resolution in semantic recognition as claimed in claim 1, characterized in that: In the said Step 4, convert each time noun into absolute time, while considering the relative time influence in the front and back time nouns. If a certain time entity has an impact on the front or back time noun, then this entity needs to be aligned to the corresponding time noun.

Images