A historical document feature analysis method and system based on logical inference rules

Abstract

The application discloses a historical literature feature analysis method and system based on logical reasoning rules, and relates to the technical field of text processing; the method acquires historical literature texts to be analyzed and divides them into multiple narrative windows, determines target narrative windows containing at least two official appointment anchors and official position recurrence words; for any target narrative window, target official position recurrence words located after the official appointment anchors are identified, candidate continuation relationships thereof relative to each official appointment anchor are generated; then, according to official position name matching values, text distance values, intra-sentence subject-object relationship matching values, official position hierarchical relationship matching values, character continuity marking values and official appointment relationship interruption marking values, a continuation priority is determined, a target official appointment anchor is determined, and an analysis relationship between the official position recurrence words, the target official appointment anchor and a character is output. The application is suitable for the structured analysis of historical literature such as historical books, local chronicles and stele inscription collation texts.

Description

Technical Field

[0001] This application relates to the field of text processing technology, and in particular to a method and system for analyzing the features of historical documents based on logical reasoning rules. Background Technology

[0002] The digitization of historical documents typically requires analyzing features such as people, official titles, time, place, and events from historical records, local chronicles, inscriptions, genealogies, and memorials. Existing technologies can identify entities such as names of people, official titles, and places in historical documents through named entity recognition, dictionary matching, rule matching, or deep learning models. This provides a data foundation for historical knowledge graph construction, biographical compilation, event relationship extraction, and digital humanities research.

[0003] However, in historical documents, once a person is appointed to an official position, the name is often not repeated in subsequent texts; instead, the official title is used directly to indicate the person's connection to the preceding text. For example, a document might record, "Wang was appointed Prefect, and Li was appointed Chief Clerk. The Prefect ordered the Chief Clerk to repair the dike." In this expression, while "Prefect" and "Chief Clerk" are literal expressions of official titles, in the current narrative context they refer to Wang and Li, who have already assumed their positions. If the system only recognizes "Prefect" and "Chief Clerk" as ordinary official titles without further identifying the corresponding individuals, it will lead to errors in parsing the event subject, the issuing authority, or the executing authority.

[0004] Furthermore, when multiple official appointment anchors exist within the same narrative window, the selection of the corresponding person for the recurring official title becomes more complex. For example, if "Wang was appointed governor" and "Li was appointed chief secretary" appear simultaneously in the preceding text, and "the governor ordered the chief secretary to repair the dike" appears later, the person corresponding to the recurring official title cannot be determined simply by the most recent person, the most recent official title, or a single string match. Instead, it requires a comprehensive judgment based on factors such as the matching relationship of official titles, the hierarchical relationship of official titles, the agent-patient relationship within the sentence, text distance, the continuity of the person, and whether the appointment relationship is interrupted. Existing entity recognition methods usually focus on recognizing the official title words themselves, and existing referential resolution methods usually focus on the anaphoric relationship of pronouns or common noun phrases, making it difficult to accurately connect and resolve recurring official title anchors when multiple official appointment anchors coexist in historical documents.

[0005] Therefore, a historical document feature analysis method based on logical reasoning rules is needed to generate candidate succession relationships and determine priorities for multiple official appointment anchors and subsequent recurring official titles in the same narrative window. This would improve the accuracy of analyzing the figures corresponding to recurring official titles in historical documents and reduce the situation of incorrect connection between figure nodes and recurring official titles during the structured processing of historical documents. Summary of the Invention

[0006] This application provides a method and system for analyzing the features of historical documents based on logical reasoning rules, in order to improve the accuracy of analyzing the corresponding figures of repeated official titles in historical documents when multiple official anchor points coexist.

[0007] Firstly, this application provides a method for analyzing historical document features based on logical reasoning rules, which can be executed by a historical document feature analysis device. In this method, the historical document feature analysis device acquires the text of the historical document to be analyzed and performs sentence segmentation on the text to obtain multiple narrative windows. The device then determines at least one target narrative window from these multiple narrative windows. The target narrative window is a narrative window containing at least two official appointment anchors and a recurring word for an official title following these anchors. The official appointment anchors include character descriptions, descriptions of official relationships, descriptions of official titles, and anchor point positions. For any target narrative window, the device determines a target recurring word for an official title located after the at least two official appointment anchors. This target recurring word is related to... At least one of the at least two official position anchors must satisfy the candidate matching condition for the expression of an official position; the historical document feature analysis device generates a candidate succession relationship between the target official position recurring word and the at least two official position anchors based on the candidate matching relationship between the target official position recurring word and the at least two official position anchors; the historical document feature analysis device determines the target official position anchor from the at least two official position anchors based on the succession priority of the candidate succession relationship; the historical document feature analysis device identifies the person expression in the target official position anchor as the person referred to by the target official position recurring word, and outputs the analysis relationship between the target official position recurring word, the target official position anchor, and the person.

[0008] Using the above methods, the historical document feature analysis device does not simply identify repeated expressions of official titles as ordinary official title entities. Instead, when there are at least two official title anchors in the same narrative window, it further generates candidate succession relationships between repeated official title words and different official title anchors, and determines the actual person referred to by the repeated official title word based on the succession priority. This avoids errors in character relationship analysis caused by matching only the most recent person or a single official title word in narratives with multiple official title anchors.

[0009] In one possible design, the process of acquiring the historical document text to be parsed and segmenting it into multiple narrative windows includes: segmenting the historical document text into multiple text sentences; and dividing the multiple text sentences into multiple narrative windows based on at least one of paragraph boundaries, sentence distance, and narrative switching expressions; wherein the narrative switching expressions include at least one of expressions for a new event, a new character, a new time, and a new location. This design allows for limiting the context for comparison before formally judging the continuity of repeated official titles, avoiding erroneous continuity caused by cross-paragraph, cross-event, or cross-character narratives.

[0010] In one possible design, determining at least one target narrative window from the plurality of narrative windows includes: identifying candidate expressions for character, candidate expressions for official relationship, and candidate expressions for official rank from each narrative window; when the same text statement or a combination of adjacent text statements contains candidate expressions for character, candidate expressions for official relationship, and candidate expressions for official rank, generating official rank anchors based on the candidate expressions for character, candidate expressions for official relationship, and candidate expressions for official rank; when more than two official rank anchors are generated in any narrative window, and an official rank expression that satisfies the candidate matching condition with the official rank expression in at least one official rank anchor is identified after the two or more official rank anchors, the narrative window is determined as the target narrative window. This design can transform the "character-official relationship-official rank" structure in historical documents into computable official rank anchors, and only narrative windows with genuine competition for multiple official rank anchors and recurring official rank terms are subsequently processed, reducing misjudgments caused by irrelevant text.

[0011] In one possible design, the candidate expressions for official appointment relationships include at least one of the following: granting of official position, appointment by appointment, removal from office, transfer to another post, leadership of office, acting appointment, and concurrent appointment. Generating the official appointment anchor point based on the candidate expressions for individuals, official appointment relationships, and official titles includes: recording the individual identifier corresponding to the candidate expression for an individual, the official title identifier corresponding to the candidate expression for an official title, the type of official appointment relationship corresponding to the candidate expression for an official appointment relationship, the text statement containing the official appointment anchor point, and the position of the anchor point, thereby generating the official appointment anchor point. Through this design, the official appointment anchor point not only includes individuals and official titles but also records the type and position of the official appointment relationship, enabling the subsequent determination of the priority of repeated words related to official titles by combining the official appointment type and text distance.

[0012] In one possible design, determining the recurring term of the target official title located after the at least two official title anchors within the target narrative window includes: identifying official title expressions within a preset sentence range following the at least two official title anchors; when the official title expression satisfies at least one of the following criteria—name matching, abbreviation matching, alias matching, or preset general term mapping matching—the official title expression is determined as the target recurring term of the official title; wherein, the preset general term mapping matching refers to the existence of a directly recorded general term mapping relationship between the official title expression and the official title expression in the official title anchors in a preset official title rule base, the directly recorded general term mapping relationship includes the general term expression, the mappable official title identifier, and the mapping application conditions, and does not include matching relationships formed solely due to the same official title category. This design allows for locating recurring terms of official titles that may follow previous figures after the official title anchors, and allows matching through abbreviations, aliases, and preset general term mappings, adapting to situations where official title expressions in historical documents are not entirely consistent.

[0013] In one possible design, generating candidate succession relationships between the target official title recurring word and the at least two official position anchors based on the candidate matching relationships between the target official title recurring word and the at least two official position anchors includes: determining whether the target official title recurring word and the official title expression in each official position anchor satisfy the candidate matching conditions; when the target official title recurring word and the official title expression in any official position anchor satisfy the candidate matching conditions, generating a candidate succession relationship between the target official title recurring word and that official position anchor; wherein, the candidate succession relationship includes the target official title recurring word, the person expression in the official position anchor, the official title expression in the official position anchor, the text distance between the target official title recurring word and the official position anchor, and the matching type. Through this design, the problem of "which official position recurring word might correspond to" can be transformed into a comparison problem of candidate succession relationships, giving the parsing process a clear intermediate data structure.

[0014] In one possible design, determining the target official anchor point from at least two official anchor points based on the priority of the candidate continuation relationships includes: determining the priority of each candidate continuation relationship based on the official title matching value, text distance value, and intra-sentence agent-patient relationship matching value; and further determining the priority of each candidate continuation relationship based on the official title hierarchy relationship matching value when there are more than two recurring official title words in the target sentence containing the target official title recurring word; wherein, the priority of each candidate continuation relationship is also adjusted based on at least one of the person continuity marker value and the official relationship interruption marker value; and the official anchor point corresponding to the candidate continuation relationship with the highest continuation priority is determined as the target official anchor point. This design avoids mechanical continuation based solely on text distance or identical official titles, and instead comprehensively considers factors such as official title hierarchy, intra-sentence semantics, person continuity, and interruption status to determine the target official anchor point.

[0015] In one possible design, when a target statement containing the target official title recurring word contains two or more official title recurring words, and the semantic type of the governing verb in the target statement is a command-type action verb, the intra-sentence agent-patient relationship matching value of the candidate succession relationship corresponding to the official title recurring word located before the governing verb and with a higher official title level than the other official title recurring word is set as the first matching value; the intra-sentence agent-patient relationship matching value of the candidate succession relationship corresponding to the official title recurring word located after the governing verb and associated with an execution-type action verb is set as the second matching value; the succession priority of the corresponding candidate succession relationship is adjusted based on the first matching value and the second matching value. Through this design, when two or more official title recurring words appear simultaneously in the target statement, the person each official title recurring word belongs to can be determined by combining intra-sentence relationships such as "who issues the order, who receives the order, and who executes the order," avoiding multiple official title recurring words being incorrectly assigned to the same official position anchor.

[0016] In one possible design, before determining the target official anchor point from the at least two official anchor points based on the priority of the candidate succession relationships, the method further includes: detecting whether there is an interruption expression in the official relationship between the official anchor point corresponding to the candidate succession relationship and the target official title recurring word; when the interruption expression is a death expression, dismissal expression, or resignation expression, prohibiting the official anchor point corresponding to the candidate succession relationship from being determined as the target official anchor point; when the interruption expression is a transfer expression, recall expression, or demotion expression, setting the official relationship interruption flag value of the candidate succession relationship to a first flag value, and reducing the succession priority of the candidate succession relationship based on the first flag value. Through this design, the succession relationship of the original official anchor point can be blocked when the official position of a person has changed, preventing the system from incorrectly parsing subsequent official title recurring words as a substitute for a person who has been transferred, resigned, or died.

[0017] Secondly, this application provides a historical document feature analysis system based on logical reasoning rules. The system includes a text acquisition module, an anchor point determination module, a recurring word determination module, a sequence relationship generation module, a target anchor point determination module, and a character analysis module. The text acquisition module acquires the text of the historical document to be analyzed and segments it into multiple narrative windows. The anchor point determination module determines at least one target narrative window from the multiple narrative windows. The recurring word determination module, for any target narrative window, determines the at least two official anchor points located within that target narrative window. The system then generates candidate succession relationships for the target official title recurring words based on the candidate matching relationships between the target official title recurring words and the at least two official title anchor points. The target anchor point determination module determines the target official title anchor point from the at least two official title anchor points based on the succession priority of the candidate succession relationships. The character analysis module identifies the character representation in the target official title anchor point as the character referred to by the target official title recurring words and outputs the analysis relationship between the target official title recurring words, the target official title anchor point, and the character. Through this system, different functional modules can collaboratively implement the above method, forming a closed-loop processing flow for official title anchor point identification, official title recurring word location, candidate succession relationship generation, and character analysis relationship output in historical documents.

[0018] In summary, this application addresses the difficulty in accurately identifying the corresponding figures for recurring official titles in historical documents when multiple official titles coexist. By structuring the expression of the figure, the expression of the official relationship, the expression of the official title, and the anchor position into official title anchors, and generating candidate succession relationships for subsequent recurring official titles, it avoids erroneous succession caused by relying solely on the nearest person, identical strings, or ordinary entity recognition. By introducing intra-sentence agent-patient relationships, official hierarchical relationships, text distance, figure continuity, and interruption markers of official relationships for comprehensive judgment, it can distinguish the issuing subject and the receiving or executing subject in scenarios such as the governor ordering the chief clerk to repair the dike, reducing erroneous connections between figure nodes and event nodes, and improving the accuracy, consistency, and verifiability of historical document knowledge graphs, figure resume compilation, and event relationship extraction results. Attached Figure Description

[0019] Figure 1 A flowchart illustrating a method for parsing historical document features based on logical reasoning rules, provided in this application embodiment; Figure 2 A schematic diagram of a narrative window and an official anchor point provided in an embodiment of this application; Figure 3 This is a schematic diagram illustrating the generation of a candidate succession relationship, provided in an embodiment of this application. Figure 4 This is a schematic diagram illustrating the determination of acceptance priority in an embodiment of this application; Figure 5 A schematic diagram of the structure of a historical document feature parsing system based on logical reasoning rules provided in this application embodiment; Figure 6 This is a possible structural diagram of a historical document feature parsing device provided in an embodiment of this application. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the embodiments of this application will be further described in detail below with reference to the accompanying drawings; wherein, in the description of the embodiments of this application, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly indicating the number of indicated technical features.

[0021] The historical document feature parsing method based on logical reasoning rules provided in this application embodiment can be executed by a historical document feature parsing device; the historical document feature parsing device can be a server, a terminal device, or a text processing platform implemented collaboratively by a server and a terminal device; the historical document feature parsing device can be used to acquire the historical document text to be parsed, and to parse the expressions of characters, official titles, official relationship expressions, and repeated words of official titles in the historical document text to be parsed, so as to output the parsing relationship between repeated words of official titles and the characters they refer to.

[0022] This application does not merely provide rules for manual reading, but rather uses a historical document feature analysis device to automatically segment sentences, generate structured anchor points, construct candidate succession relationships, and calculate succession priorities in the historical document text to be analyzed. It outputs parsed relationship data that can be stored, retrieved, and called by a computer, thereby reducing the occurrence of incorrect connections between recurring official titles and person nodes during the construction of historical document knowledge graphs, and improving the consistency and verifiability of structured text processing results.

[0023] In this application embodiment, the historical document text to be parsed can be a historical document text containing expressions of the relationship between people holding official positions, the succession of official positions, or the recurrence of official positions; for example, the historical document text to be parsed can be a history book, local chronicle, inscription text, genealogy text, memorial text, or other text containing narratives of people holding official positions; this application embodiment does not limit the specific source and storage format of the historical document text to be parsed, and the historical document text to be parsed can be manually entered text, text obtained after optical character recognition, or structured or semi-structured text stored in a database.

[0024] In historical documents, once a person is appointed to a certain official position, the official position is often directly used in the following text to indicate the person mentioned earlier. For example, the document may contain the phrase "Wang was appointed as governor, and Li was appointed as chief secretary; the governor ordered the chief secretary to repair the dike." In this example, "governor" and "chief secretary" in the following text are not just ordinary official titles, but refer to Wang and Li, who have already taken office in the previous text. If only the official position entity recognition is performed, although the system can identify "governor" and "chief secretary" as official titles, it cannot accurately determine the corresponding person, which can easily lead to errors in the subject relationship of historical events.

[0025] Based on the above problems, this application provides a method for analyzing the features of historical documents based on logical reasoning rules; such as Figure 2 As shown, in this method, the historical document feature parsing device first divides the historical document text to be parsed into multiple narrative windows, and determines at least one target narrative window containing at least two official position anchors and recurring official titles from among the multiple narrative windows; then, for any target narrative window, it determines the target official title recurring word; next, it generates candidate succession relationships between the target official title recurring word and different official position anchors; finally, it determines the target official position anchor based on the succession priority of the candidate succession relationships, and identifies the person expressed in the target official position anchor as the person referred to by the target official title recurring word; thus, it can accurately parse the person corresponding to the official title recurring word in historical document narratives with multiple official position anchors.

[0026] Figure 1 A flowchart illustrating a historical document feature parsing method based on logical reasoning rules is provided for embodiments of this application; see reference. Figure 1 The method includes the following steps: S101: The historical document feature analysis device acquires the text of the historical document to be analyzed, and divides the text of the historical document to be analyzed into sentences to obtain multiple narrative windows; Among them, the historical document feature analysis device can perform sentence segmentation on the historical document text to be analyzed to obtain multiple text sentences; the sentence segmentation can be achieved based on periods, semicolons, line breaks, paragraph marks, ancient book collation punctuation, or preset sentence segmentation models; a text sentence refers to the smallest narrative processing unit obtained after sentence segmentation.

[0027] Historical document feature analysis equipment can divide multiple text statements into multiple narrative windows based on at least one of paragraph boundaries, sentence distance, and narrative switching expressions. A narrative window refers to a set of text statements that revolve around the same or related people, events, times, or places within a preset range. Narrative switching expressions can include at least one of the following: expressions of a new event, a new person, a new time, and a new place. For example, when expressions such as "afterwards," "otherwise," "someone else," or "next year in a certain place" appear between adjacent text statements, indicating a clear shift in the narrative object, time, or place, the historical document feature analysis equipment can use that position as the boundary of the narrative window.

[0028] In one implementation, the historical document feature analysis device can initially divide consecutive text statements in the same paragraph into a narrative window; if the number of text statements in the narrative window exceeds a preset sentence spacing threshold, it can be further divided according to changes in the expression of the person or time; if there is the same person or official position in adjacent paragraphs, the adjacent paragraphs can also be merged into a narrative window; the preset sentence spacing threshold can be preset by the system or configured according to the type of historical document, for example, set to three sentences, five sentences or eight sentences.

[0029] It should be noted that the narrative switching expression is used to divide the narrative window, while the subsequent official relationship interruption expression is used to reduce or block the candidate succession relationship. The two have different functions. The narrative switching expression is mainly used to determine whether the text scope has entered a new narrative unit; the official relationship interruption expression is mainly used to determine whether a certain official anchor point can still serve as the source of succession for subsequent official position recurrence words.

[0030] S102: The historical document feature parsing device determines at least one target narrative window from the plurality of narrative windows; The target narrative window is a narrative window containing at least two official appointment anchors and a repeated word of official title after the at least two official appointment anchors; the official appointment anchors include character expression, official appointment relationship expression, official title expression, and anchor position; the character expression can be the name, courtesy name, pen name, title, or other expression that can point to the character in historical documents; the official title expression can be the expression that indicates the official title, official status, or appointment title in historical documents; the official appointment relationship expression can be the expression that indicates the relationship of conferment of official title, appointment, transfer, concurrent appointment, or acting appointment; the anchor position can be the position of the text sentence containing the official appointment anchor, the character position, the sentence order position, or the paragraph position.

[0031] The historical document feature parsing device can identify person candidate expressions, official appointment relationship candidate expressions, and official position candidate expressions from each narrative window; when the same text sentence or a combination of adjacent text sentences contains a person candidate expression, an official appointment relationship candidate expression, and an official position candidate expression, the historical document feature parsing device can generate an official appointment anchor point based on the person candidate expression, the official appointment relationship candidate expression, and the official position candidate expression; when more than two official appointment anchor points are generated in any narrative window, and an official position expression that meets the candidate matching condition with the official position expression in at least one official appointment anchor point is identified after the more than two official appointment anchor points, the narrative window is determined as a target narrative window; The official appointment relationship candidate expressions may include at least one of an official granting expression, an appointment expression, a formal appointment expression, a promotion appointment expression, a concurrent acting expression, an acting appointment expression, and a concurrent appointment expression; for example, words such as "serve as", "appoint", "formally appoint", "grant", "promote", "concurrently act as", "act as", and "concurrently hold" can be used as the official appointment relationship candidate expressions; when the historical document feature parsing device generates an official appointment anchor point, it may record the person identifier corresponding to the person candidate expression, the official position identifier corresponding to the official position candidate expression, the official appointment relationship type corresponding to the official appointment relationship candidate expression, the text sentence where the official appointment anchor point is located, and the anchor point position; For example, the target narrative window includes the text sentence "Wang serves as the provincial governor, Li serves as the chief clerk", and the historical document feature parsing device can generate two official appointment anchor points; wherein the first official appointment anchor point includes the person expression "Wang", the official appointment relationship expression "serve as", the official position expression "provincial governor" and the corresponding anchor point position; the second official appointment anchor point includes the person expression "Li", the official appointment relationship expression "serve as", the official position expression "chief clerk" and the corresponding anchor point position; when "the provincial governor orders the chief clerk to build a dyke" appears in the subsequent text, this narrative window satisfies both the condition of more than two official appointment anchor points and the condition of the presence of the official position recurring word, so it can be determined as the target narrative window.

[0032] In an implementation, the person candidate expressions, official position candidate expressions, and official appointment relationship candidate expressions can be identified through rule dictionary identification, named entity recognition model identification, or a combination of rule dictionary and named entity recognition model identification; the embodiments of the present application do not limit the specific recognition model, as long as it can output person candidate expressions, official position candidate expressions, and official appointment relationship candidate expressions.

[0033] It should be noted that a to-be-parsed historical document text may have multiple target narrative windows; the historical document feature parsing device may perform subsequent steps for each target narrative window respectively, or perform subsequent steps sequentially according to the occurrence order of the target narrative windows in the to-be-parsed historical document text.

[0034] S103: For any target narrative window, the historical document feature parsing device determines a target official position recurring word located after the at least two official appointment anchor points in the target narrative window; Wherein the target official position recurrence term is an official position expression that satisfies a candidate matching condition with at least one official position expression in at least two official appointment anchors; the historical document feature parsing device may identify an official position expression within a preset statement range after the at least two official appointment anchors; when the identified official position expression satisfies at least one of same-name matching, short name matching, alternative name matching or preset generic name mapping matching with at least one official position expression in the at least two official appointment anchors, the identified official position expression is determined as the target official position recurrence term.

[0035] The same-name matching means that the target official position recurrence term and the official position expression in the official appointment anchor have the same text or are identical after standardization; for example, "Cishi" (provincial governor) and "Cishi" satisfy the same-name matching; the short name matching means that the target official position recurrence term is a short name of the official position expression in the official appointment anchor or an expression obtained by omitting the administrative region and rank modifiers; for example, "Jiangzhou Cishi" (Jiangzhou provincial governor) and "Cishi" satisfy the short name matching; the alternative name matching means that the target official position recurrence term and the official position expression in the official appointment anchor are recorded as an alternative name relationship in the preset official position rule base; the preset generic name mapping matching means that there is a directly recorded generic name mapping relationship between the target official position recurrence term and the official position expression in the official appointment anchor in the preset official position rule base, and the directly recorded generic name mapping relationship includes a generic name expression, a mappable official position identifier and a mapping application condition, and does not include a matching relationship formed only because the official positions belong to the same category.

[0036] Optionally, a generic name mapping identifier, a generic name expression, a mappable official position identifier, a mapping application period, a mapping application condition and a mapping priority may be set in the preset official position rule base; for example, the generic name expression "Junshou" (prefectural governor) may be recorded in the preset official position rule base as being mappable to the official position identifier "Taishou" (grand guardian), with the application condition that no other official position of the same-level local chief appears in the target statement; if it is only recorded that both "Cishi" and "Zhangshi" (chief clerk) belong to the local official category, the preset generic name mapping matching is not constituted; accordingly, the boundary of the preset generic name mapping matching may be limited, and excessive merging between different official positions may be avoided.

[0037] It should be noted that the superior-subordinate relationship between official positions is not directly used as a candidate matching condition for the target official position recurrence term; the superior-subordinate official position relationship is mainly used for subsequent adjustment of the inheritance priority; for example, there may be a superior-subordinate relationship between "Cishi" and "Zhangshi", but "Zhangshi" should not be determined as a recurrence term of the official appointment anchor of "Cishi" only because it has a superior-subordinate relationship with "Cishi"; accordingly, it is possible to avoid incorrectly merging different persons corresponding to different official positions; For example, in the sentence "Wang Mou is Jiangzhou Cishi; the Cishi arrives at the prefecture", the subsequent "Cishi" satisfies the short name matching with the preceding "Jiangzhou Cishi", so "Cishi" may be determined as the target official position recurrence term; for another example, in the sentence "Wang Mou is Cishi, Li Mou is Zhangshi; the Cishi orders the Zhangshi to build a dyke", the subsequent "Cishi" and "Zhangshi" respectively satisfy the same-name matching with the official position expressions in the preceding official appointment anchors, so both may be used as official position recurrence terms.

[0038] As Figure 3 shown, S104: the historical document feature parsing device generates a candidate inheritance relationship of the target official position reproduction word relative to the at least two official appointment anchors according to the candidate matching relationship between the target official position reproduction word and the at least two official appointment anchors; Specifically, the historical document feature parsing device may respectively determine whether the target official position reproduction word and the official position expression in each official appointment anchor satisfy a candidate matching condition; when the target official position reproduction word and the official position expression in any official appointment anchor satisfy the candidate matching condition, generate a candidate inheritance relationship between the target official position reproduction word and this official appointment anchor; the candidate inheritance relationship may include the target official position reproduction word, the person expression in this official appointment anchor, the official position expression in this official appointment anchor, the text distance between the target official position reproduction word and this official appointment anchor, and the matching type.

[0039] In an embodiment, the candidate inheritance relationship may be represented by the following structure: R={W, H, O, Dist, M}; where R denotes the candidate inheritance relationship, W denotes the target official position reproduction word, H denotes the person expression in the candidate official appointment anchor, O denotes the official position expression in the candidate official appointment anchor, Dist denotes the text distance between the target official position reproduction word and the candidate official appointment anchor, M denotes the matching type; the text distance may be the number of statements spaced between the text statement where the target official position reproduction word is located and the text statement where the candidate official appointment anchor is located, or may be the character distance or paragraph distance between the two. For example, the target official position reproduction word is "Cishi (Provincial Governor)", the official position expression in the first official appointment anchor is "Cishi", and the official position expression in the second official appointment anchor is "Zhangshi (Chief Clerk)", then the historical document feature parsing device may generate the candidate inheritance relationship of "Cishi - first official appointment anchor"; for another example, the target official position reproduction word is "Junshou (Prefectural Governor)", and when it is recorded in the preset official position rule base that "Junshou" can be directly mapped to "Taishou (Prefectural Governor)" in the official appointment anchor, the candidate inheritance relationship of "Junshou - Taishou official appointment anchor" may be generated.

[0040] As Figure 4 shown, S105: the historical document feature parsing device determines a target official appointment anchor from the at least two official appointment anchors according to the inheritance priority of the candidate inheritance relationship; In one implementation, the historical document feature analysis device can determine the priority of each candidate succession relationship based on the matching value of official titles, text distance value, and intra-sentence agent-patient relationship matching value. When there are more than two recurring words of official titles in the target sentence containing the recurring words of the target official title, the device can further determine the priority of each candidate succession relationship by combining the matching value of official title hierarchy relationship. The historical document feature analysis device can also adjust the priority of each candidate succession relationship based on at least one of the continuity marker value of the person and the interruption marker value of the official appointment relationship, and then determine the official appointment anchor point corresponding to the candidate succession relationship with the highest succession priority as the target official appointment anchor point.

[0041] Optionally, for any candidate succession relationship, a succession priority value Q can be calculated: Q = aN + bL + cD + dS + eC - fI; where Q is the succession priority value, N is the official title matching value, L is the official title hierarchy matching value, D is the text distance value, S is the intra-sentence agent-patient relationship matching value, C is the person continuity marker value, I is the official relationship interruption marker value, and a, b, c, d, e, and f are preset weights and are all non-negative numbers; the values ​​of N, L, D, S, C, and I can range from 0 to 1; a, b, c, d, e, and f can be preset by the system, manually configured, or adjusted based on labeled samples, and can be normalized so that a + b + c + d + e + f = 1; when no normalization is performed, the historical document feature analysis device can perform relative sorting according to the size of the Q value of each candidate succession relationship; For example, in an embodiment where the primary goal is to accurately connect repeated official titles, values ​​such as a=0.30, b=0.15, c=0.15, d=0.25, e=0.10, and f=0.30 can be set. Among these, the matching values ​​of official titles and the matching values ​​of agent-patient relationships within sentences have higher weights, while the value of the interruption marker of official relationship has a higher weight as a deduction item. The above values ​​are only examples and can be adjusted according to the labeled samples or different types of historical documents.

[0042] The official title matching value N is used to represent the degree of matching between the target official title recurring word and the official title expressed in the candidate official title anchor point; for example, N is 1 when matching the same name, N is 0.8 when matching the abbreviation or alternative name, and N is 0.6 when matching the default general title mapping; the above values ​​are only examples and can be adjusted according to the type of historical documents and the rule base configuration in specific implementation.

[0043] The official rank relationship matching value L is used to indicate whether the official rank in the candidate official anchor point matches the intra-sentence role in the target sentence; for example, in "the governor ordered the chief clerk to repair the dike", "order" is a command-type action word, and "governor" is on the commanding side. If the official rank of "governor" is higher than that of "chief clerk" in the preset official rank rule library, then the L value of the candidate succession relationship corresponding to "governor" can be 1; if the official rank does not match the intra-sentence role, then L can be 0 or a lower value.

[0044] The text distance value D is used to represent the distance influence between the target official position recurrence word and the candidate official-appointment anchor; optionally, D=1 / (1+m), where m is the number of sentences spaced between the sentence where the target official position recurrence word is located and the sentence where the candidate official-appointment anchor is located; the smaller the number of spaced sentences, the higher the text distance value.

[0045] The intra-sentence agent-patient relationship matching value S is used to indicate whether the semantic role assumed by the target official position recurrence word in the target sentence matches the official position attribute of the candidate official-appointment anchor; the intra-sentence agent-patient relationship may include an order-issuing role, an order-receiving role, and an executing role; the historical document feature parsing device may determine the semantic type of the governing verb according to a preset action word list, and determine the intra-sentence agent-patient relationship corresponding to the official position recurrence word according to the position of the official position recurrence word being before the governing verb, after the governing verb, after a preposition, or after a serial verb construction.

[0046] The person continuity mark value C is used to indicate whether the person in the candidate official-appointment anchor is still in the current main narrative line; the historical document feature parsing device may generate the person continuity mark value according to whether a new explicit person subject appears between the candidate official-appointment anchor and the target official position recurrence word, whether a new official-appointment anchor appears, whether a person switching expression appears, and whether an official-appointment relationship interruption expression appears; optionally, if no competing explicit person subject appears after the candidate official-appointment anchor, C is set to 1; if a new explicit person subject appears but no official-appointment relationship interruption is constituted, C is set to 0.5; if an official-appointment relationship interruption expression such as transfer of post, removal from office, or death appears, C is set to 0.

[0047] The official-appointment relationship interruption mark value I is used to indicate whether there is an official-appointment relationship interruption expression between the candidate official-appointment anchor and the target official position recurrence word; if there is an official-appointment relationship interruption expression, I is set to 1; if there is no official-appointment relationship interruption expression, I is set to 0; in the above formula, I is used as a deduction item, which can reduce the inheritance priority of the candidate inheritance relationship with an interruption situation.

[0048] The preset official position rule base may include at least one field of an official position identifier, a standard official position name, a short official position name, an alternative official position name, an official position category, an affiliated dynasty or valid period, an official position level, a superior official position identifier, a subordinate official position identifier, a common general term, a general term mapping identifier, a general term expression, a mappable official position identifier, a mapping applicable period, a mapping applicable condition, and a mapping priority; for example, the preset official position rule base may record an official position identifier O001, a standard official position name "Cishi", a short official position name "Cishi", an official position category "local chief executive", an official position level "superior", and a subordinate official position identifier O002; and may further record an official position identifier O002, a standard official position name "Zhangshi", an official position category "assistant official", an official position level "subordinate", and a superior official position identifier O001; The preset action word list may include at least one of the following fields: action word, semantic type of action word, agent role position, patient role position, execution role position, whether to trigger subject inheritance, and whether to trigger interruption judgment; for example, the action word "ming" (order) may be recorded in the preset action word list, with its semantic type of action word being an order-issuing action word, the agent role position being before the action word, and the patient role position being after the action word; the action word "xiu" (repair) may also be recorded, with its semantic type of action word being an execution-type action word, and the execution role position being before the action word or inheriting the previous command-receiving role; the preset action word list may further record non-inheritance action words, for example, narrative action words such as "yue" (say), "yun" (state), and "zai" (record) do not directly trigger the agent-patient role judgment of the official position reappearance word. When the target sentence only contains narrative action words, the historical document feature analysis device may reduce the matching value of the intra-sentence agent-patient relationship, and instead preferentially determine the candidate inheritance relationship according to the official position name matching value and the text distance value; For example, in the sentence "Wang Mou serves as the governor, Li Mou serves as the chief clerk; the governor orders the chief clerk to repair the embankment", the dominant verb in the target sentence is "order", "governor" is located on the order-issuing side, "chief clerk" is located on the order-receiving side or the execution side, and there is a superior-subordinate relationship between "governor" and "chief clerk" in the preset official position rule base; therefore, the historical document feature analysis device may inherit "governor" to the official appointment anchor point including "Wang Mou — governor", and inherit "chief clerk" to the official appointment anchor point including "Li Mou — chief clerk".

[0049] In another embodiment, when there are more than two official position reappearance words in the target sentence containing the target official position reappearance word, the historical document feature analysis device may generate a corresponding candidate inheritance relationship for each official position reappearance word respectively; then set the intra-sentence agent-patient relationship matching value of each candidate inheritance relationship according to the arrangement positions of the more than two official position reappearance words in the target sentence, the semantic type of the dominant verb in the target sentence, and the official position hierarchy relationship corresponding to the more than two official position reappearance words, and adjust the inheritance priority of each candidate inheritance relationship based on the intra-sentence agent-patient relationship matching value; the semantic type of the dominant verb may be determined according to the preset action word list, which includes at least one of order-issuing action words, order-receiving action words, appointment action words, governance action words, execution-type action words, and non-inheritance action words.

[0050] For example, the target sentence is "The governor orders the chief clerk to repair the embankment"; the historical document feature analysis device identifies two official position reappearance words "governor" and "chief clerk"; the dominant verb "order" belongs to the order-issuing action word, so "governor" located before "order" is identified as the order-issuing role, and "chief clerk" located after "order" and associated with "repair the embankment" is identified as the order-receiving role or the execution role; if the official position hierarchy of "governor" in the preset official position rule base is higher than that of "chief clerk", the priority of the candidate inheritance relationship between "governor" and the official appointment anchor point "Wang Mou — governor" is increased, and the priority of the candidate inheritance relationship between "chief clerk" and the official appointment anchor point "Li Mou — chief clerk" is increased; For another example, the target sentence is "the Chief Clerk receives the order from the Regional Inspector to build the dyke". Although "Chief Clerk" is at the beginning of the sentence, due to the presence of the "receive...order" structure in the target sentence, the historical document feature analysis device can identify "Chief Clerk" as an order-receiving role or an executing role and identify "Regional Inspector" as an order-issuing role according to a preset action word list; For still another example, the target sentence is "the Regional Inspector orders to build the dyke", there is only one official title recurring word "Regional Inspector" in the target sentence, and "order" is an order-issuing action word, so the historical document feature analysis device can preferentially parse "Regional Inspector" as the order-issuing subject; for another example, the target sentence is "the Chief Clerk builds the dyke", in which "build" is an execution-type action word, and "Chief Clerk" is located before the action word, so the historical document feature analysis device can parse "Chief Clerk" as the executing subject.

[0051] In yet another implementation, before determining the target official appointment anchor point, the historical document feature analysis device may further detect whether there is an official appointment relationship interruption expression between the official appointment anchor point corresponding to the candidate inheritance relationship and the target official title recurring word; when the official appointment relationship interruption expression is a death expression, a removal expression or a departure expression, the historical document feature analysis device may prohibit determining the official appointment anchor point corresponding to the candidate inheritance relationship as the target official appointment anchor point; when the official appointment relationship interruption expression is a transfer expression, a recall expression or a demotion expression, the historical document feature analysis device may set the official appointment relationship interruption flag value I of the candidate inheritance relationship to a first flag value, and reduce the inheritance priority of the candidate inheritance relationship based on the first flag value; the first flag value may be 1, and is used as a deduction item in the inheritance priority formula.

[0052] For example, in the context of "Wang is the Regional Inspector, and soon transfers to be the Prefect; the Regional Inspector orders officials to build the dyke", "transfer to be the Prefect" can serve as an interruption expression of the original "Regional Inspector" official appointment relationship, the historical document feature analysis device can set the official appointment relationship interruption flag value I of the candidate inheritance relationship corresponding to the "Wang - Regional Inspector" official appointment anchor point to 1, and reduce the inheritance priority based on this flag value, or prohibit taking this official appointment anchor point as the target official appointment anchor point for the subsequent "Regional Inspector"; for another example, in the context of "Wang is the Regional Inspector, and soon transfers to be the Prefect; the Prefect arrives at the prefecture to relieve the people", the subsequent "Prefect" matches the "Wang - Prefect" official appointment anchor point formed after the transfer, so "Prefect" can be inherited to the new official appointment anchor point after the transfer.

[0053] In a further embodiment, when the undertaking priority values of more than two candidate undertaking relationships are the same, the historical document feature analysis device may determine the target official appointment anchor according to at least one of matching type priority, text distance, sequence of anchor positions, and manual review mark; wherein, the matching type priority may be set such that same-name matching is higher than short name matching, short name matching is higher than alias matching, and alias matching is higher than preset general term mapping matching; if the target official appointment anchor still cannot be determined, the historical document feature analysis device may output a plurality of candidate analysis relationships, and mark the plurality of candidate analysis relationships as a to-be-reviewed state.

[0054] S106: The historical document feature analysis device determines the person expression in the target official appointment anchor as the person referred to by the target official post recurrence word, and outputs the analysis relationship among the target official post recurrence word, the target official appointment anchor and the person; For example, in the text "Wang is the provincial governor, Li is the chief clerk; the provincial governor orders the chief clerk to repair the dyke", the historical document feature analysis device may output the analysis relationship of the provincial governor, the first official appointment anchor, and Wang, as well as the analysis relationship of the chief clerk, the second official appointment anchor, and Li; Accordingly, when subsequently constructing a person activity trajectory, event subject relationship or historical knowledge graph, the issuing subject of "ordering dyke repair" can be analyzed as Wang, and the executing subject or instructed subject of "dyke repair" can be analyzed as Li.

[0055] In an embodiment, the analysis relationship may include at least one field of an official post recurrence word identifier, a target official appointment anchor identifier, a person identifier, an official post identifier, an undertaking priority, a matching type, and a source position; Through the analysis relationship, the intermediate basis for why the official post recurrence word is parsed as a certain person can be retained, which facilitates subsequent review, display or knowledge graph storage.

[0056] The processing manner of the embodiment of the present application is described below with reference to specific examples: In the first example, the to-be-analyzed historical document text includes "Wang is the provincial governor, Li is the chief clerk; the provincial governor orders the chief clerk to repair the dyke"; the historical document feature analysis device generates two official appointment anchors of "Wang — provincial governor" and "Li — chief clerk", and identifies two official post recurrence words of "provincial governor" and "chief clerk" in the subsequent text; According to the commanding semantics of the dominant verb "order", the arrangement positions of the two official post recurrence words and the official post hierarchy relationship, the historical document feature analysis device determines that "provincial governor" undertakes the official appointment anchor of "Wang — provincial governor", and "chief clerk" undertakes the official appointment anchor of "Li — chief clerk".

[0057] Example 2: The historical document text to be analyzed includes "Wang was the Prefect of Jiangzhou; the Prefect arrived in the prefecture and repaired the dike"; the historical document feature analysis device identifies "Prefect of Jiangzhou" as the official title expression in the appointment anchor point, and "Prefect" is a recurring word of the official title in the following text; since "Prefect" is the abbreviation of "Prefect of Jiangzhou", the two satisfy the abbreviation matching, so the historical document feature analysis device can generate a candidate succession relationship of "Prefect - Wang - Prefect of Jiangzhou" and identify Wang as the person referred to by "Prefect".

[0058] Example 3: The historical document text to be analyzed includes "Wang was appointed governor, and later promoted to prefect; the governor ordered officials to repair the dike"; the historical document feature analysis device identifies the "Wang - governor" appointment anchor point, but it identifies the interruption of the appointment relationship between this appointment anchor point and the "governor" in the following text. Therefore, it lowers the priority of the "Wang - governor" appointment anchor point, or prohibits it from being the target appointment anchor point of the "governor" in the following text, thereby avoiding the subsequent "governor" being incorrectly analyzed as Wang.

[0059] Example 4: The historical document text to be analyzed includes "Wang was appointed governor; later Li arrived in the prefecture; the governor repaired the dike." The historical document feature analysis device can identify a new explicit person, "Li." However, if "Li arrived in the prefecture" does not form a new official appointment anchor point, nor does it form an expression of the interruption of the official appointment relationship between "Wang and governor," then the historical document feature analysis device can set the continuity marker value of the person to an intermediate value, such as 0.5, and then combine the official title matching value, text distance value, and other matching values ​​to comprehensively determine the target official appointment anchor point of "governor" in the following text.

[0060] Example 5: The historical document text to be parsed includes "Zhang was appointed Prefect; the Prefect arrived at the prefecture and opened the granaries to provide relief to the people"; the preset official title rule library directly records that "Prefect" can be mapped to "Prefect" under this document type, and no other local official titles of the same level appear in the target sentence. Therefore, "Prefect" and "Prefect" satisfy the preset general title mapping match; the historical document feature parsing device can generate a candidate succession relationship of "Prefect - Zhang - Prefect" and identify Zhang as the person referred to by "Prefect".

[0061] Example 6, the to-be-parsed historical document text includes "Wang is the Cishi, Li is the Zhangshi; the Cishi orders the Zhangshi to build the dyke"; in one weight configuration, for the candidate inheritance relationship of "Cishi - Wang - Cishi", N=1, L=1, D=0.5, S=1, C=1, I=0, and a=0.30, b=0.15, c=0.15, d=0.25, e=0.10, f=0.30, then Q=0.30×1+0.15×1+0.15×0.5+0.25×1+0.10×1-0.30×0=0.875; if the Q value of another candidate inheritance relationship is lower than 0.875, the historical document feature parsing device determines the office-holding anchor point "Wang - Cishi" as the target office-holding anchor point of "Cishi".

[0062] Example 7, the to-be-parsed historical document text includes "Wang is the Cishi, Li is the Zhangshi; historical records show that the former Cishi's administrative office was located north of the city, and the Zhangshi's official residence is next to it", where "record" is a non-inheritance action word, which does not directly trigger the judgment of the commanding role or the commanded role; the historical document feature parsing device can reduce the matching value of the agent-patient relationship within the sentence, mainly determine the candidate inheritance relationships of "Cishi" and "Zhangshi" respectively based on the official position name matching value and the text distance value, or output a to-be-reviewed mark when the scores of the candidate inheritance relationships are close.

[0063] Example 8, the first paragraph of the to-be-parsed historical document text includes "Wang is the Cishi, Li is the Zhangshi; the Cishi orders the Zhangshi to build the dyke"; the second paragraph includes "Zhao is the Taishou, Qian is the Xianling; the Taishou orders the Xianling to relieve the people"; the historical document feature parsing device can determine the first paragraph and the second paragraph as two target narrative windows respectively, and execute official position recurring word inheritance parsing respectively, so as to obtain two sets of mutually non-contaminated parsing relationships in the same document.

[0064] Figure 5 is a schematic structural diagram of a historical document feature parsing system based on logical inference rules provided by the embodiments of the present application; reference Figure 5 , the system may include a text acquisition module, an anchor point determination module, a recurring word determination module, an inheritance relationship generation module, a target anchor point determination module, and a person parsing module: The text acquisition module acquires the historical document text to be parsed and segments it into sentences to obtain multiple narrative windows. The anchor point determination module determines at least one target narrative window from the multiple narrative windows. The target narrative window is a narrative window containing at least two official appointment anchor points and a recurring word of official title after the at least two official appointment anchor points. The official appointment anchor points include character descriptions, descriptions of official relationships, descriptions of official titles, and anchor point positions. The recurring word determination module determines, for any target narrative window, a target official title recurring word located after at least two official appointment anchor points within the target narrative window. The current word is an official title expression that satisfies the candidate matching condition with at least one of the at least two official title anchors; the succession relationship generation module is used to generate candidate succession relationships between the target official title recurring word and at least two official title anchors based on the candidate matching relationships between the target official title recurring word and at least two official title anchors; the target anchor point determination module is used to determine the target official title anchor point from at least two official title anchor points based on the succession priority of the candidate succession relationships; the person parsing module is used to determine the person expression in the target official title anchor point as the person referred to by the target official title recurring word, and output the parsing relationship between the target official title recurring word, the target official title anchor point, and the person.

[0065] It is understood that the module division in this application embodiment is illustrative and is only a logical functional division. In actual implementation, there may be other division methods. Each functional module can be integrated into a processor, or it can exist physically separately, or two or more modules can be integrated into one module. The integrated module can be implemented in hardware or as a software functional module.

[0066] Figure 6 A possible structural diagram of a historical document feature parsing device provided in an embodiment of this application; the historical document feature parsing device may include a processor, a memory, and a communication interface; The processor can be used to execute the processing steps in the above-mentioned historical document feature parsing method; The memory can be used to store the historical document text to be parsed, the preset official title rule base, the preset action word list, candidate succession relationships, succession priorities, and parsing relationships; The communication interface can be used to receive historical document text to be parsed, or to output parsing relationships to external devices.

[0067] This application may also provide a computer-readable storage medium storing a computer program, which, when executed by a computer, causes the computer to perform the methods described in the embodiments of this application; this application may also provide a computer program product, which, when run on a computer, causes the computer to perform the methods described in the embodiments of this application.

[0068] Those skilled in the art should understand that the above embodiments can be implemented by software, hardware, or a combination of software and hardware; the methods, steps, and modules disclosed in the embodiments of this application can be implemented using general-purpose processors, digital signal processors, application-specific integrated circuits, field-programmable gate arrays, or other programmable logic devices; the above embodiments are only used to illustrate the technical solutions of this application and are not intended to limit the scope of protection of this application; any modifications, equivalent substitutions, or improvements made within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. A method for analyzing the features of historical documents based on logical reasoning rules, characterized in that, The method, applied to a historical document feature analysis device, includes: Obtain the text of the historical document to be parsed, and divide the text into sentences to obtain multiple narrative windows; At least one target narrative window is determined from the plurality of narrative windows. The target narrative window is a narrative window that contains at least two official appointment anchors and has a recurring word of official title after the at least two official appointment anchors. The official appointment anchors include character expression, official relationship expression, official title expression and anchor position. For any target narrative window, determine the target official title recurring word located after the at least two official title anchors in the target narrative window. The target official title recurring word is an official title expression that satisfies the candidate matching condition with at least one of the official title expressions in the at least two official title anchors. Based on the candidate matching relationship between the target official title recurring word and the at least two official appointment anchors, generate the candidate succession relationship of the target official title recurring word relative to the at least two official appointment anchors, including: determining whether the official title expression in each official appointment anchor satisfies the candidate matching condition. When the target official title recurring word and the official title expression in any official anchor point satisfy the candidate matching condition, a candidate succession relationship between the target official title recurring word and the official anchor point is generated. The candidate succession relationship includes the target official title recurring words, the character expressions in the official title anchor point, the official title expressions in the official title anchor point, the text distance between the target official title recurring words and the official title anchor point, and the matching type. Based on the priority of the candidate succession relationship, the target official anchor point is determined from the at least two official anchor points, including: determining the priority of each candidate succession relationship based on the official title name matching value, text distance value and intra-sentence agent-patient relationship matching value; and when there are more than two official title recurring words in the target sentence containing the target official title recurring words, the priority of each candidate succession relationship is further determined by combining the official title hierarchy relationship matching value. In addition, the priority of each candidate succession relationship is adjusted based on at least one of the character continuity marker value and the official relationship interruption marker value. The appointment anchor point corresponding to the candidate appointment relationship with the highest priority is determined as the target appointment anchor point. The official rank hierarchy matching value is determined based on a preset official rank rule base, and the intra-sentence agent-patient relationship matching value is determined based on the governing verb in the text sentence containing the target official rank recurring word and the position of the target official rank recurring word relative to the governing verb. The person expressed in the target official position anchor point is identified as the person referred to by the target official position recurring word, and the parsing relationship between the target official position recurring word, the target official position anchor point and the person is output.

2. The method according to claim 1, characterized in that, The process involves acquiring the historical document text to be parsed, segmenting the text into sentences, and obtaining multiple narrative windows, including: The historical document text to be parsed is segmented into multiple text sentences; The plurality of text statements are divided into the plurality of narrative windows based on at least one of paragraph boundaries, sentence distance, and narrative switching expression; The narrative switching expression includes at least one of the following: expression of a new event, expression of a new character, expression of a new time, and expression of a new location.

3. The method according to claim 1, characterized in that, Determining at least one target narrative window from the plurality of narrative windows includes: Identify candidate expressions for characters, official relationships, and official titles within each narrative window; When the same text statement or a combination of adjacent text statements contains candidate expressions for people, candidate expressions for official relationships, and candidate expressions for official positions, an official position anchor point is generated based on the candidate expressions for people, candidate expressions for official relationships, and candidate expressions for official positions. When two or more official appointment anchors are generated in any narrative window, and an official appointment expression that meets the candidate matching condition with the official appointment expression in at least one official appointment anchor is identified after the two or more official appointment anchors, the narrative window is determined as the target narrative window.

4. The method according to claim 3, characterized in that, The candidate expressions for official appointment relationships include at least one of the following: expressions for conferring official position, appointing to office, removing from office, transferring to another position, leading to office, acting to office, and concurrent to office. The step of generating appointment anchor points based on the candidate expressions for individuals, the candidate expressions for official relationships, and the candidate expressions for official positions includes: Record the person identifier corresponding to the candidate expression of the person, the official position identifier corresponding to the candidate expression of the official position, the official position relationship type corresponding to the candidate expression of the official position relationship, the text statement where the official position anchor point is located, and the position of the anchor point, and generate the official position anchor point.

5. The method according to claim 1, characterized in that, The step of determining the target official title recurring words located after the at least two official anchor points in the target narrative window includes: Identify official titles within a preset sentence range following the at least two official appointment anchors; When the expression of official title satisfies at least one of the following conditions with at least one of the at least two official title anchor points: same name matching, abbreviation matching, alternative name matching, or preset general term mapping matching, the expression of official title is determined as the target official title recurring word; The preset general term mapping matching refers to the existence of a directly recorded general term mapping relationship between the official title expression and the official title expression in the official appointment anchor point in the preset official title rule base. The directly recorded general term mapping relationship includes the general term expression, the mappable official title identifier and the mapping applicable conditions, and does not include the matching relationship formed only because the official title category is the same.

6. The method according to claim 1, characterized in that, The process of determining the priority of each candidate continuation relationship based on the matching value of official titles, text distance value, and intra-sentence agent-patient relationship matching value, and further determining the priority of each candidate continuation relationship by combining the matching value of official title hierarchy relationship when there are more than two recurring official titles in the target sentence containing the target official title, includes: When there are two or more repeated official titles in the target statement containing the target official title repeated words, and the semantic type of the governing verb in the target statement is a commanding action verb, the intra-sentence agent-patient relationship matching value of the candidate succession relationship corresponding to the official title repeated word that is located before the governing verb and whose official title level is higher than that of another official title repeated word is set as the first matching value. Set the intra-sentence agent-patient relationship matching value of the candidate succession relationship corresponding to the official title recurrence word located after the dominating verb and associated with the execution action word as the second matching value; The acceptance priority of the corresponding candidate acceptance relationship is adjusted based on the first matching value and the second matching value; The semantic type of the governing verb is determined according to a preset action word list, which includes at least one of the following: commanding action words, receiving action words, appointing action words, governing action words, executing action words, and non-continuing action words.

7. The method according to any one of claims 1 to 6, characterized in that, Before determining the target appointment anchor point from the at least two appointment anchor points based on the appointment priority of the candidate appointment relationship, the method further includes: Detect whether there is an interruption in the expression of the official relationship between the anchor point of the candidate succession relationship and the recurring word of the target official position; When the interruption of the appointment relationship is expressed as death, dismissal, or resignation, it is prohibited to determine the appointment anchor point corresponding to the candidate succession relationship as the target appointment anchor point. When the interruption of the appointment relationship is expressed as a transfer, recall, or demotion, the interruption flag value of the appointment relationship of the candidate successor relationship is set to the first flag value, and the succession priority of the candidate successor relationship is reduced based on the first flag value.

8. A historical document feature parsing system based on logical reasoning rules, used to implement the method described in any one of claims 1-7, characterized in that, include: The text acquisition module is used to acquire the text of the historical document to be parsed, and to divide the text of the historical document to be parsed into sentences to obtain multiple narrative windows; An anchor point determination module is used to determine at least one target narrative window from the plurality of narrative windows. The target narrative window is a narrative window that contains at least two official appointment anchor points and has recurring words of official titles after the at least two official appointment anchor points. The official appointment anchor points include character expression, official relationship expression, official title expression, and anchor point position. The recurrence word determination module is used to determine, for any target narrative window, the target official title recurrence word located after the at least two official title anchors in the target narrative window, wherein the target official title recurrence word is an official title expression that satisfies the candidate matching condition with at least one of the official title expressions in the at least two official title anchors. The succession relationship generation module is used to generate a candidate succession relationship between the target official title recurring word and the at least two official appointment anchors based on the candidate matching relationship between the target official title recurring word and the at least two official appointment anchors. The target anchor point determination module is used to determine the target official anchor point from the at least two official anchor points according to the succession priority of the candidate succession relationship. The succession priority is determined based on the official title matching value, the text distance value, and the intra-sentence agent-patient relationship matching value. The character analysis module is used to identify the character expression in the target official anchor point as the character referred to by the target official position recurring word, and output the analysis relationship between the target official position recurring word, the target official anchor point and the character.

Images