A Method for Building Expert Profiles for Blind Review of Graduate Theses

Abstract

This invention discloses a method for constructing expert profiles for blind review of graduate theses, specifically in the field of blind review data processing, to address the disconnect between academic matching and review behavior in expert selection. The method involves acquiring submitted thesis data and expert historical data, merging the expert historical data by identity, standardizing the format, and aligning the time to form a basic profile record. Based on the time alignment results, a review behavior trajectory is constructed, identifying the scoring deviation position and the turning point of the scoring deviation. Combining the lag in review time and the fixation of the scoring deviation direction, an ordinal cross-result is formed to obtain an expert behavior profile. Semantic representation processing is performed on the submitted thesis data and expert academic information to obtain a candidate expert set. The expert behavior profile is mapped to the expert academic representation to form a candidate ranking result, generating a blind review expert list. The actual review results are then reflected back to the basic profile record to maintain or update the expert behavior profile.

Description

Technical Field

[0001] This invention relates to the field of blind review data processing, and more specifically, to a method for constructing expert profiles for blind review scenarios of graduate theses. Background Technology

[0002] In blind review scenarios for graduate theses, submitting institutions typically manage submitted thesis data, expert historical data, and review results through a blind review business system. Current blind review expert selection methods often rely on matching experts by discipline, research direction, keywords, or expert database tags, combined with business criteria such as professional title, region, number of reviews received, and conflict of interest avoidance to screen candidate experts. While this method can complete basic screening, it primarily depends on static academic information and struggles to reflect changes in expert scoring deviations, review time variations, and whether review behavior has evolved over time in consecutive blind review tasks.

[0003] In practice, whether an expert is suitable for inclusion in the blind review expert list depends not only on whether the expert's academic information covers the research direction of the submitted papers, but also on the stability of the expert's historical review process. For example, the same expert may give scores that are too high, too low, or too close at different stages, and the review time may also shift forward, remain the same, or shift backward depending on the review task. Without establishing expert profiles, it is impossible to unify the expert's basic attributes, academic information, historical review process, review behavior trajectory, score deviation position, and review time position under the same expert index, and it is also impossible to use academic matching results and review behavior status simultaneously when ranking candidates.

[0004] Meanwhile, historical expert data may also have issues such as missing expert identifiers, experts with the same name, inconsistent spelling of affiliated institutions, inconsistent expression of discipline names, and missing review or submission dates. Furthermore, there may be discrepancies between submitted paper data and expert academic information, such as empty research direction fields, inconsistencies in hierarchical or synonymous descriptions. Therefore, it is necessary to establish expert profiles for blind review of graduate theses, enabling the submission paper representation, expert academic representation, and expert behavioral profiles to jointly participate in the acquisition of candidate expert sets, the formation of candidate ranking results, and the reflection of actual review outcomes.

[0005] To address the aforementioned problems, a technical solution is provided. Summary of the Invention

[0006] To overcome the aforementioned deficiencies of the prior art, embodiments of the present invention provide a method for constructing expert profiles for blind review scenarios of graduate theses. This method involves acquiring submitted thesis data and expert historical data, merging the expert historical data by identity, standardizing the format, and aligning the time to form a basic profile record; constructing a review behavior trajectory based on the time alignment results, determining the scoring deviation position and the turning point of the scoring deviation, and combining the review time lag and the fixed amount of the scoring deviation direction to form an ordinal cross result, thus obtaining an expert behavior profile; performing semantic representation processing on the submitted thesis data and expert academic information to obtain a candidate expert set; mapping the expert behavior profile to the expert academic representation to form a candidate ranking result, generating a blind review expert list, and reflecting the actual review results back to the basic profile record to maintain or update the expert behavior profile, thereby solving the problems mentioned in the background art.

[0007] To achieve the above objectives, the present invention provides the following technical solution:

[0008] S1. Obtain the submitted paper data and expert historical data from the blind review business system, merge the expert historical data by identity, unify the format and align the time, and form a basic record of profiles with experts as the index.

[0009] S2. Based on the time alignment results in the basic portrait record, arrange the historical review process of each expert into a review behavior trajectory. Determine the turning point of the score deviation according to the position of the score deviation in the review behavior trajectory, and use the turning point of the score deviation as the boundary to form the review behavior segment.

[0010] S3. Calculate the review time transition lag and the score deviation direction fixation for adjacent review behavior segments. Based on the review time transition lag and the score deviation direction fixation, form the ordinal cross result. When the ordinal cross result points to the conflict segment, recalculate the time alignment of the profile basic record. Otherwise, determine the review behavior segment after the score deviation transition segment as the expert behavior profile.

[0011] S4. Perform semantic representation processing on the submitted paper data and the expert academic information in the profile basic record to form the submitted paper representation and the expert academic representation. Obtain the candidate expert set based on the similarity between the submitted paper representation and the expert academic representation.

[0012] S5. Map the expert behavior profiles in the candidate expert set to the corresponding expert academic representations to form the candidate ranking results. Based on the candidate ranking results, form a blind review expert list and reflect the actual review results corresponding to the blind review expert list back to the profile base record. When the actual review results form a new conflict segment in the review behavior trajectory, update the boundary of the turning point of the score deviation; otherwise, maintain the expert behavior profile.

[0013] Furthermore, identity merging includes: when expert identifiers exist and are consistent, the corresponding historical review processes are grouped into the same expert index; when at least one historical review process lacks an expert identifier, the expert name, institution, and discipline name are first processed before merging, and then, when the expert name, institution, and discipline name are consistent, they are grouped into the same expert index; when any field is inconsistent, a record to be merged is formed; the format is standardized by organizing the historical review process into fields such as submitted paper data identifier, review time, submission time, submission deadline, and scoring result, and the completeness of the record is formed based on the existence of the fields.

[0014] Furthermore, the scoring deviation position is determined as follows: Based on the data identifier of the submitted papers, a set of review results corresponding to the same submitted paper is obtained. When the set of review results contains scoring results outside the current trajectory position, a directional comparison is performed. When the set of review results lacks comparable review results, a set of review results from the same batch and research direction is obtained. When neither of the two sets of review results is sufficient to form a control position, the current trajectory position is recorded as an incomparable scoring position. When a control position is obtained, the scoring result is recorded as higher, lower, or aligned with the control position based on its relative height or position. A turning point in the scoring deviation is formed when a directional shift occurs between higher and lower, and the shifted direction continues in subsequent adjacent trajectory positions.

[0015] Furthermore, the review time transition lag is formed as follows: when both the acceptance time, submission time, and submission deadline exist, and the submission deadline is later than the acceptance time, the progress position of the submission time within the review cycle of the submission task formed by the acceptance time and the submission deadline is determined as the review time position; if the acceptance time, submission time, or submission deadline is missing, the corresponding trajectory position does not form a review time position; when the review time position changes from moving forward or remaining flat to moving backward and the backward movement continues to appear in the next adjacent trajectory position, a review time transition position is formed, and the ordinal cross result is formed only when the review time transition lag and the score deviation direction fixation amount correspond to the same score deviation transition segment.

[0016] Furthermore, the main research direction in the submitted paper is determined as follows: when the research direction field exists, the direction description in the research direction field is determined as the main research direction; when the research direction field is empty, the paper title is matched with the direction name, discipline name, and research object name in the research direction directory. If the paper title matches an entry in the research direction directory, the main research direction is determined. If the paper title does not match an entry in the research direction directory, the submitted paper does not form a main research direction and is not included in the candidate expert set for processing; if the direction description in the expert's academic statement is consistent with the main research direction and belongs to a subordinate direction of the main research direction, or if the discipline description in the expert's academic statement covers the discipline scope of the main research direction and the direction description corresponds to the research object, it is determined to cover the main research direction.

[0017] Furthermore, when there are incomparable scoring positions in the review behavior trajectory, these incomparable scoring positions are not included in the confirmation of scoring deviation turning points and are retained in the review behavior trajectory. When there are no scoring deviation turning points in the review behavior trajectory, the entire review behavior trajectory forms a non-turning-point review behavior segment. This segment is not divided into preceding and subsequent review behavior segments and is processed as a non-turning-point review behavior segment. When all scoring deviation positions in the non-turning-point review behavior segment are aligned and the review time consumption positions do not form a subsequent shift, the non-turning-point review behavior segment is identified as an expert behavior profile. When the review time consumption positions in the non-turning-point review behavior segment form a subsequent shift, an expert behavior profile is not formed and is retained as a review behavior segment pending verification.

[0018] Furthermore, when forming the blind review expert list, if the number of expert indices in the candidate ranking results is less than the required number of reviews, the blind review expert list is formed according to the existing candidate ranking results, and a supplementary review mark is retained in the profile basic record; when the actual review result is inserted at the end of the review behavior trajectory and there is no subsequent trajectory position, if the review time position of the actual review result is shifted to the previous trajectory position, the trajectory position of the actual review result is recorded as the review time turning point to be confirmed and no new conflict segment is formed; after the next historical review process enters the review behavior trajectory of the same expert index, local verification continues to be performed based on the trajectory position of the actual review result, the subsequent trajectory position, and the scoring deviation position.

[0019] Furthermore, the basic profile records include the basic attributes of experts under the same expert index, expert academic information, historical review processes with pre-ordered trajectories, records to be merged, and verification records; historical review processes with pre-ordered trajectories retain the data identifier of the submitted paper, the time of acceptance, the time of submission, the deadline for submission, and the scoring results; records to be merged retain the original expert's name, institution, and discipline; verification records retain duplicate historical review processes that did not participate in time alignment; when both the time of acceptance and the time of submission are missing, the corresponding historical review process is retained as an unsorted record.

[0020] Furthermore, each candidate expert in the candidate expert set retains an expert index, expert academic representation, and an entry point for the expert behavior profile corresponding to the expert index; each candidate record in the candidate ranking results retains an expert index, expert academic representation, source of the expert behavior profile, and its ranking position; the blind review expert list obtains the expert index according to the ranking position in the candidate ranking results, and retains the expert academic representation, source of the expert behavior profile, and its ranking position in the candidate ranking results; when the number of expert indexes in the candidate ranking results is less than the required number of reviews, a supplementary review mark is retained in the profile base record.

[0021] Furthermore, the actual review results include expert index, submitted paper data identifier, review time, submission time, submission deadline, and scoring results. After the actual review results are reflected back to the basic profile record, the landing point of the actual review results in the review behavior trajectory is determined according to the review time or submission time. If both the review time and submission time are missing, the actual review results are retained as unsorted records. After the location of the actual review results is recorded as the turning point of the review time to be confirmed, when the next historical review process enters the review behavior trajectory and forms a backward continuation, it is first determined whether a new turning point of scoring deviation has been formed within the local verification scope. Then, a new ordinal cross result is formed based on the new review time turning point lag and the new scoring deviation direction fixation. When the new ordinal cross result points to the conflict segment, the boundary of the turning point of scoring deviation is updated.

[0022] The technical effects and advantages of the expert profiling method for blind review of graduate theses in this invention are as follows:

[0023] This invention merges expert identifiers, expert names, affiliated institutions, and discipline names, and combines this with pre-merging field organization, record completeness, and time alignment processing to ensure that historical expert data forms a consistent profile record based on the expert index. The unified organization of fields such as submitted paper data identifiers, review acceptance time, submission time, submission deadline, and scoring results provides a clear data source for scoring deviations, review time consumption, and review behavior trajectories, reducing profile bias caused by missing fields or duplicate records.

[0024] This invention processes scoring deviations and review time consumption within the same review behavior trajectory. It uses the turning points of scoring deviations, the lag in review time transitions, and the fixation of scoring deviation direction to form a sequence cross-result. This ensures that expert behavior profiles not only rely on static academic information but also reflect changes in direction and time consumption throughout the historical review process. Conflict segments only trigger time alignment recalculation within their corresponding range, avoiding a complete recalculation of the entire profile's basic records and controlling the scope of data rearrangement.

[0025] This invention simultaneously uses submitted papers, expert academic representations, and expert behavioral profiles when forming the candidate expert set, and reflects these profiles back to the baseline record after the actual review results are generated. The scoring deviations and review time consumption in the actual review results can be partially verified by tracing the original review behavior trajectory. This allows the expert behavioral profiles to be maintained or updated according to the new blind review process, reducing the instability issues arising from relying solely on academic text matching for candidate expert ranking. Attached Figure Description

[0026] Figure 1 This is a flowchart illustrating the expert profile construction method for blind review of graduate theses according to the present invention.

[0027] Figure 2 This is a schematic diagram of the image-based record formation process of the present invention;

[0028] Figure 3 This is a schematic diagram illustrating the turning point between the review behavior trajectory and the score deviation in this invention.

[0029] Figure 4 This is a schematic diagram illustrating the formation of the submitted paper and candidate expert set for this invention. Detailed Implementation

[0030] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0031] Please see Figures 1-4 This invention provides a method for constructing expert profiles for blind review scenarios of graduate theses, including:

[0032] S1. Obtain the submitted paper data and expert historical data from the blind review business system, merge the expert historical data by identity, unify the format and align the time, and form a basic record of profiles with experts as the index.

[0033] S2. Based on the time alignment results in the basic portrait record, arrange the historical review process of each expert into a review behavior trajectory. Determine the turning point of the score deviation according to the position of the score deviation in the review behavior trajectory, and use the turning point of the score deviation as the boundary to form the review behavior segment.

[0034] S3. Calculate the review time transition lag and the score deviation direction fixation for adjacent review behavior segments. Based on the review time transition lag and the score deviation direction fixation, form the ordinal cross result. When the ordinal cross result points to the conflict segment, recalculate the time alignment of the profile basic record. Otherwise, determine the review behavior segment after the score deviation transition segment as the expert behavior profile.

[0035] S4. Perform semantic representation processing on the submitted paper data and the expert academic information in the profile basic record to form the submitted paper representation and the expert academic representation. Obtain the candidate expert set based on the similarity between the submitted paper representation and the expert academic representation.

[0036] S5. Map the expert behavior profiles in the candidate expert set to the corresponding expert academic representations to form the candidate ranking results. Based on the candidate ranking results, form a blind review expert list and reflect the actual review results corresponding to the blind review expert list back to the profile base record. When the actual review results form a new conflict segment in the review behavior trajectory, update the boundary of the turning point of the score deviation; otherwise, maintain the expert behavior profile.

[0037] This invention focuses on processing submitted paper data and expert historical data. First, the expert historical data is merged by identity, standardized in format, and aligned by time, so that historical review processes are aggregated under the corresponding expert index and form a basic profile record. Based on this, historical review processes with pre-ordered trajectories are arranged into review behavior trajectories. The position of score deviation is determined by comparing the direction of the scoring results relative to the set of review results, and the turning point and review behavior segment of the score deviation are determined based on the directional changes in the continuous trajectory positions. Subsequently, the review time position and score deviation position are read around adjacent review behavior segments to form the review time turning point lag, the score deviation direction fixation, and the order cross result, distinguishing conflict segments and expert behavior profiles. Next, the submitted paper data and expert academic information are organized into submitted paper representations and expert academic representations, and a candidate expert set is obtained based on the main research direction and auxiliary direction. Finally, the expert behavior profiles in the candidate expert set are mapped to the corresponding expert academic representations to form the candidate ranking results and the blind review expert list. The actual review results are reflected back to the basic profile record, new conflict segments are judged, and the turning point boundary of the score deviation is updated.

[0038] Detailed implementation of step S1:

[0039] The blind review system in this invention refers to a data processing platform used by submitting institutions to manage the blind review process of graduate theses. This system stores or retrieves at least the submitted thesis data, expert historical data, review task configurations, and actual review results. The submitted thesis data includes the thesis data identifier, thesis title, research direction, and keywords; expert historical data includes basic expert attributes, expert academic information, and historical review processes; review task configurations include submission deadlines; and actual review results include expert indexes, submitted thesis data identifiers, acceptance time, submission time, submission deadline, and scoring results. The blind review system can be a self-built blind review management platform of the submitting institution or a third-party blind review management platform used by the submitting institution. As long as the above data sources can be provided, it falls under the category of the blind review system defined in this invention. Upon entering S1, the blind review system already possesses submitted paper data and expert historical data. Submitted paper data includes data identifiers, titles, research directions, and keywords. Expert historical data includes basic expert attributes, academic information, and historical review processes. Historical review processes include expert identifiers, submitted paper data identifiers, acceptance time, submission time, submission deadline, and scoring results. The expert index is formed through identity merging and is used to aggregate historical review processes belonging to the same expert. The submission deadline originates from the submission task configuration in the blind review system; the period from acceptance time to submission deadline constitutes the review cycle of the submission task. The processing goal of S1 is to organize the scattered expert historical data into a profile-based record indexed by the expert. However, expert historical data often suffers from issues such as missing expert identifiers, inconsistent field formats, and incomplete time sequences. Therefore, it needs to be organized according to the order of identity merging, format standardization, and time alignment.

[0040] S101 Identity Merging Processing.

[0041] S101 retrieves historical review processes from the blind review system's historical expert data. Each historical review process is treated as a pending record, containing the expert identifier, expert name, institution, discipline, and corresponding review content. Identity merging begins with processing the expert identifier, which distinguishes the expert's identity and typically originates from the expert account number or expert database number in the blind review system. For any two historical review processes, if both have identical expert identifiers, they are grouped into the same expert index; otherwise, they are not grouped into the same expert index.

[0042] When at least one historical review process lacks an expert identifier, the identity merging process continues by reading the expert's name, affiliation, and discipline. The expert's name identifies the individual, the affiliation identifies the source of the expert's position, and the discipline identifies the scope of the expert's field. Only when the expert's name, affiliation, and discipline all match will the corresponding historical review process be grouped into the same expert index; if any of these fields is inconsistent, the corresponding historical review process will be retained as a record to be merged. Records to be merged are not included in time alignment and retain their original field content to avoid erroneous merging due to experts with the same name, changes in affiliation, or cross-disciplinary appointments.

[0043] Before comparing expert names, affiliations, and discipline names, the fields for these information are first processed to ensure consistency. This pre-merge field processing includes removing leading and trailing spaces, standardizing full-width and half-width characters, standardizing the full and abbreviations of the affiliation, and standardizing the format of the discipline name in the blind review system's discipline directory. The full and abbreviations of the affiliation are derived from the affiliation name table in the blind review system, and the standard format of the discipline name is derived from the discipline directory in the blind review system. After completing the pre-merge field processing, it is then determined whether the expert name, affiliation, and discipline name are all consistent.

[0044] In one embodiment, historical review process A has an expert identifier U001, and historical review process B also has an expert identifier U001; both are directly included in expert index E001. Historical review process C lacks an expert identifier, but the expert's name, institution, and subject name are consistent with existing records in expert index E001; historical review process C is included in expert index E001. Historical review process D lacks an expert identifier, and the expert's name is consistent with existing records in expert index E001, but the institution is different; historical review process D is retained as a record to be merged and is not merged into expert index E001. Through the above processing, the same expert index only carries historical review processes with clear identity sources or consistent field combinations.

[0045] S102 format is processed uniformly.

[0046] S102 processes historical expert data based on the expert index formed in S101. Under each expert index, the basic attributes of the experts are organized into fields for expert name, institution, discipline, professional title, and region. Expert academic information is organized into fields for discipline name and research direction. Historical review processes are organized into fields for submitted paper data identifier, review acceptance time, submission time, submission deadline, and scoring result. When organizing the fields, the date format is standardized to the same time format, the scoring result is standardized to a scoring expression recognizable by the blind review system, the research direction is standardized to the same textual definition as the research direction in the submitted paper data, and the submission deadline is standardized to the deadline expression in the blind review system's submission task configuration.

[0047] When both the acceptance time and submission time exist in the historical review process, the review time is calculated by subtracting the acceptance time from the submission time, with time as the unit of measurement. If either the acceptance time or the submission time is missing, the review time is not calculated, and a time-missing marker is retained in the historical review process. The time-missing marker is only used to indicate that the historical review process lacks the necessary fields for calculating the review time and does not change the identity merging result of the expert index.

[0048] For historical review processes with duplicate content under the same expert index, S102 selects those participating in time alignment based on record completeness. Record completeness represents the number of valid fields within a historical review process. Valid fields include the submitted paper data identifier, acceptance time, submission time, submission deadline, and scoring result. The record completeness increases by one item if the submitted paper data identifier exists, the record completeness increases by one item if the acceptance time exists, the record completeness increases by one item if the submission time exists, the record completeness increases by one item if the submission deadline exists, and the record completeness increases by one item if the scoring result exists. If the record completeness differs between duplicate historical review processes, the historical review process with higher record completeness is retained for time alignment; the duplicate historical review process with lower record completeness is used as a verification record. If the record completeness is the same for duplicate historical review processes, historical review processes that simultaneously possess the submitted paper data identifier, acceptance time, submission time, submission deadline, and scoring result are prioritized for time alignment; other duplicate historical review processes are used as verification records. Through the above processing, data fields under the same expert index have a unified structure, and the review time, scoring deviation position, and review time position all have a clear source.

[0049] S103 Time Alignment Processing.

[0050] Based on the unified field structure obtained in S102, S103 arranges the historical review processes that participated in time alignment under the same expert index. The arrangement prioritizes the acceptance time; historical review processes with existing acceptance times are arranged in chronological order. Historical review processes with missing acceptance times but existing submission times are inserted between adjacent historical review processes according to their submission times. Historical review processes with both missing acceptance and submission times are not included in the time alignment results but are retained as unsorted records under the corresponding expert index.

[0051] When two historical review processes have the same acceptance time, the submission time is compared, with the earlier submission time ranked first. If both acceptance and submission times are the same, the record completeness is compared, with the process having higher record completeness ranked first. If acceptance, submission, and record completeness are all identical, only one historical review process is retained for time alignment, and the remaining duplicates are used as verification records. After time alignment, each participating historical review process obtains a pre-order position, which indicates its chronological position within the same expert index, and its dimension is ordinal.

[0052] Finally, the basic attributes of experts under the same expert index, expert academic information, historical review processes with pre-ordered trajectories, records to be merged, and verification records are collectively organized into a profile base record. Among these, the historical review processes with pre-ordered trajectories express the chronological order of the expert's historical review process; the records to be merged retain identity-ambiguous data that is not yet merged; and the verification records retain the source of verification for duplicate data. After the profile base record is formed, S2 can arrange the review behavior trajectories according to the time alignment results in the profile base record.

[0053] Detailed implementation of step S2:

[0054] Step S1 has organized the historical data of experts into a profile-based record indexed by the experts, and retained the historical review process with trajectory pre-sequence position in the profile-based record; however, the trajectory pre-sequence position only indicates the time sequence of the historical review process and cannot reflect the directional change of the expert scoring results relative to the results of similar blind reviews. Therefore, step S2 needs to convert the time alignment results into review behavior trajectory and determine the scoring deviation position and the turning point of the scoring deviation in the review behavior trajectory.

[0055] S201 Review Behavior Trajectory Arrangement.

[0056] S201 reads the basic profile record formed in step S1, and extracts the historical review process with pre-sequence positions from the basic profile record, using the same expert index as the processing unit. The pre-sequence positions are derived from the time alignment result of step S1, indicating the sequential position of the historical review process within the same expert index. The historical review processes are arranged from front to back according to the pre-sequence positions to form the review behavior trajectory.

[0057] Each trajectory position in the review behavior trajectory corresponds to a historical review process. Each trajectory position retains the submitted paper data identifier, scoring result, acceptance time, submission time, and review time. The submitted paper data identifier is used to obtain the review result set in S202, the scoring result is used to determine the scoring deviation position, and the acceptance time, submission time, and review time are used to maintain the time source of the historical review process. Unsorted records in the profile basic record are not included in the review behavior trajectory, and the verification records in the profile basic record are only used to verify the field source and are not considered as independent trajectory positions.

[0058] During the sorting process, if there is only one historical review process with a pre-ordered trajectory under the same expert index, a single-point review behavior trajectory is formed. This single-point trajectory only records the trajectory position and does not enter the confirmation of the turning point in the scoring deviation. If there are multiple historical review processes with pre-ordered trajectory positions under the same expert index, a multi-point review behavior trajectory is formed according to the pre-ordered trajectory positions. The multi-point review behavior trajectory is used to determine the scoring deviation position at each trajectory position in S202 and to read the directional changes of continuous trajectory positions in S203.

[0059] After S201, the time alignment results in the basic profile record are converted into review behavior trajectory. The historical review process no longer exists as scattered records, but forms a time series that can be read point by point according to the expert index and trajectory preorder.

[0060] S202 score deviation location determined.

[0061] S202 processes each trajectory position in the review behavior trajectory as the processing object, reads the submitted paper data identifier and scoring result at the trajectory position, and obtains the review result set based on the submitted paper data identifier. The review result set is used to provide a comparison object for the scoring result at the current trajectory position, and the sources of the review result set are in chronological order.

[0062] First, based on the data identifiers of the submitted papers, obtain the set of review results corresponding to the same submitted paper. If the set of review results corresponding to the same submitted paper contains review results other than those at the current trajectory position, use the set of review results corresponding to the same submitted paper for directional comparison. Second, if the set of review results corresponding to the same submitted paper lacks comparable review results, obtain the set of review results from the same batch and the same research direction. "Same batch" is used to limit the submission time, and "research direction" is used to limit the content of the paper, ensuring that the score deviation does not cross significantly different blind-reviewed objects for comparison.

[0063] The review result set is insufficient to form a control position because it does not contain any scoring results other than the current trajectory position, or the scoring results in the review result set cannot be mapped to the scoring level used by the blind review business system, or the scoring results in the review result set lack corresponding data identifiers for the submitted papers. The review result set used to form a control position does not include the scoring results for the current trajectory position itself.

[0064] If neither the set of review results for the same submitted paper nor the set of review results for the same batch and research direction is sufficient to form a reference position, then the current trajectory position is recorded as an incomparable scoring position. Incomparable scoring positions are not included in the confirmation of turning points in scoring deviation, but are still retained in the review behavior trajectory to maintain the chronological order of the historical review process.

[0065] After obtaining the review result set, the scores in the review result set are arranged from low to high according to the scoring levels used in the blind review business system. The scoring level in the middle position after arrangement is taken as the reference position. If the number of scores in the review result set is even, the business level interval between the two middle scoring levels is taken as the reference position. The business level interval is composed of the two middle scoring levels. When the score at the current trajectory position is higher than the higher scoring level in the business level interval, the current trajectory position is recorded as a high-scoring deviation position; when the score at the current trajectory position is lower than the lower scoring level in the business level interval, the current trajectory position is recorded as a low-scoring deviation position; when the score at the current trajectory position is equal to either of the two middle scoring levels, the current trajectory position is recorded as a close-fitting deviation position. When the score of the current trajectory position is higher than that of the control position, the current trajectory position is recorded as a high-scoring deviation position; when the score of the current trajectory position is lower than that of the control position, the current trajectory position is recorded as a low-scoring deviation position; when the score of the current trajectory position falls into the control position, the current trajectory position is recorded as a close-scoring deviation position.

[0066] In one embodiment, the review behavior trajectory of expert index E001 includes multiple trajectory positions, one of which corresponds to a submitted paper in the field of education management. If multiple experts provide review results for the same submitted paper, these results are arranged according to their rating levels, and the rating of E001 at the current trajectory position is compared with the ranked reference position. If only E001 provides a rating for the same submitted paper, making it impossible to make a comparison within the same paper, the review results set of papers in the same batch in the field of education management is used. If E001's rating is higher than the reference position, the current trajectory position is recorded as a high-rating deviation position; if it is lower than the reference position, the current trajectory position is recorded as a low-rating deviation position; if it falls within the reference position, the current trajectory position is recorded as a close-fitting deviation position.

[0067] Through S202, a score deviation position is obtained for each comparable trajectory position in the review behavior trajectory. The score deviation position expresses the directional state of the score result relative to the comparable blind review results, rather than expressing the magnitude of the score result itself.

[0068] The determination of the turning point of S203 score deviation and the formation of the review behavior segment.

[0069] S203 reads the score deviation position obtained from S202 and checks the directional state of continuous trajectory positions in the pre-sequence order of the review behavior trajectory. The directional state of the score deviation position includes too high, too low, and close. Too high and too low indicate that the score result is in the opposite direction to the control position, while close indicates that the score result falls into the control position.

[0070] The turning point of the scoring deviation is determined according to the directional change in the continuous trajectory position. If the scoring deviation position in the continuous trajectory position changes from a high continuous state to a low continuous state, and the low state continues to appear in subsequent adjacent trajectory positions, then the trajectory interval jointly covered by the end position of the high state, the directional change position, and the confirmation position of the continuation of the low state is determined as the turning point of the scoring deviation. If the scoring deviation position in the continuous trajectory position changes from a low continuous state to a high continuous state, and the high state continues to appear in subsequent adjacent trajectory positions, the turning point of the scoring deviation is also determined in the same way. If the scoring deviation position changes from high through a close transition to low, or from low through a close transition to high, and the opposite direction after the transition continues to appear in subsequent adjacent trajectory positions, then the trajectory interval jointly covered by the end position of the original direction, the close transition position, and the confirmation position of the continuation of the new direction is determined as the turning point of the scoring deviation.

[0071] If a single trajectory position shows a change from too high to too low or a close alignment, but the changed direction does not continue to appear in adjacent trajectory positions, then no turning point in the scoring deviation is formed. If a close alignment occurs consecutively in the review behavior trajectory, but there is no directional change between too high and too low before and after the close alignment, then no turning point in the scoring deviation is formed. Through the above processing, occasional scoring changes will not be directly identified as a stage change in the expert scoring direction.

[0072] After identifying the turning points of scoring deviation, review behavior segments are formed using these turning points as boundaries. Continuous trajectory positions before the turning point are assigned to the preceding review behavior segment, and continuous trajectory positions after the turning point are assigned to the following review behavior segment. The trajectory positions covered by the turning point serve as the boundary interval between adjacent review behavior segments. If multiple turning points exist in the review behavior trajectory, they are divided into multiple review behavior segments according to the pre-order position of the trajectory. If no turning points exist in the review behavior trajectory, the entire trajectory forms a single, non-turning-point review behavior segment. Non-turning-point review behavior segments are not divided into preceding and following review behavior segments and are used as non-turning-point input objects.

[0073] After S203, the review behavior trajectory is divided into review behavior segments with boundary intervals. The turning point of the scoring deviation can express the trajectory range of the expert scoring direction from one stage to another.

[0074] Step S2 converts the time alignment results in the basic portrait record into a review behavior trajectory, and determines the scoring deviation position, the turning point of the scoring deviation, and the review behavior segment in the review behavior trajectory, so that adjacent review behavior segments can be used as input objects for calculating the turning point lag of review time and the fixation of the scoring deviation direction in step S3; when there is no turning point of scoring deviation, the review behavior segment without turning point is used as the input object without turning point.

[0075] Detailed implementation of step S3:

[0076] Step S2 has already formed the review behavior trajectory, the turning point of the score deviation, and the adjacent review behavior segment; however, the turning point of the score deviation only reflects the stage change of the score direction, and cannot determine whether the change of review time is misaligned with the change of the score direction. Therefore, step S3 processes the review time direction and the score deviation direction in the adjacent review behavior segment respectively, and then forms the sequence cross result.

[0077] S301 Review Time Lag Calculation.

[0078] S301 takes the adjacent review behavior segments formed in step S2 as the processing object, first reads the trajectory positions in the adjacent review behavior segments, and then obtains the acceptance time, submission time, and review cycle of the submission task from the historical review process corresponding to each trajectory position. The review cycle of the submission task is the review time interval recorded by the blind review business system for the corresponding historical review process, with the starting point being the acceptance time and the ending point being the submission deadline. The review time is determined by the length of time between the acceptance time and the submission time, and the progress position of the submission time in the review cycle of the submission task is used to form the review time position.

[0079] In specific processing, first confirm whether the historical review process simultaneously has an acceptance time, a submission time, and a submission deadline. If both an acceptance time, a submission time, and a submission deadline exist, and the submission deadline is later than the acceptance time, the review time position is the progress position of the submission time within the review cycle of the submitted task. For historical review processes... The time of acceptance of the case is recorded as Submission time is recorded as The submission deadline is recorded as All three use the same unit of time and are evaluated based on the time spent. Determine according to the following formula:

[0080]

[0081] in, This refers to the review time position in the historical review process r. This refers to the length of time between the time the document is received and the time it is submitted. This refers to the length of time between the acceptance date and the submission deadline. If the submission date is later than the submission deadline, the review time position is still calculated and an overdue mark is retained. If the acceptance date, submission date, or submission deadline is missing, the current trajectory position will not be included in the review time position, and the current trajectory position will not be included in the confirmation of the review time turning point.

[0082] After the review time position is determined, the review time positions of adjacent trajectory positions are compared sequentially according to the pre-order position in the review behavior trajectory. If the review time position of the current trajectory position is greater than that of the previous trajectory position, the current trajectory position is marked as shifted forward; if the review time position of the current trajectory position is less than that of the previous trajectory position, the current trajectory position is marked as shifted forward; if the review time position of the current trajectory position is equal to that of the previous trajectory position, the current trajectory position is marked as unchanged. When the blind review business system uses working days as the smallest time unit, the review acceptance time, submission time, and submission deadline are first converted to working day order before calculating the review time position. When the review time position changes from shifted forward or unchanged to shifted backward, and the shift continues to occur in the next adjacent trajectory position, the trajectory position that first shifts backward is determined as the review time turning point. If the shift occurs only once and the next adjacent trajectory position does not continue to shift backward, a review time turning point is not formed.

[0083] After determining the turning point in review time, read the end boundary of the turning segment where the score deviates. If the turning point in review time is after the end boundary of the turning segment where the score deviates, the number of trajectory positions between the turning point in review time and the end boundary of the turning segment where the score deviates is recorded as the review time turning point lag. If the turning point in review time falls within the coverage area of ​​the turning segment where the score deviates, the review time turning point lag is recorded as an in-segment state. If the turning point in review time is before the start boundary of the turning segment where the score deviates is, the review time turning point lag is recorded as a preceding state. If no turning point in review time is formed in adjacent review behavior segments, the review time turning point lag is recorded as an unformed state.

[0084] After S301, the change in review time is converted into the trajectory sequence result around the turning point of the score deviation. The turning point lag of review time can express whether the change in review time is later than the change in the score direction.

[0085] S302 score deviation direction fixation amount calculation.

[0086] S302 continues to use the scoring deviation position and the turning point of scoring deviation formed in step S2. At the start of processing, the review behavior segment before the turning point of scoring deviation is read first. In the review behavior segment before the turning point of scoring deviation, the closest turning point of scoring deviation and the consecutive occurrence of deviations of higher or lower values ​​are found. The found higher or lower values ​​are taken as the scoring deviation direction before the turning point of scoring deviation. If the review behavior segment before the turning point of scoring deviation only contains fit, or if consecutive occurrences of higher or lower values ​​cannot be obtained, the fixation amount of the scoring deviation direction is recorded as not formed.

[0087] After obtaining the direction of the score deviation before the turning point of the score deviation, starting from the first trajectory position after the end boundary of the turning point of the score deviation, read the score deviation position according to the pre-order position of the trajectory. If the read score deviation position is the same as the direction of the score deviation before the turning point of the score deviation, the current trajectory position is determined as the score deviation continuation position. Consecutive score deviation continuation positions form a score deviation continuation segment. If the opposite direction is read, the score deviation continuation segment ends. If a fit is read, continue reading the next trajectory position; if a fit occurs consecutively after the turning point of the score deviation, and no score deviation direction before the turning point of the score deviation reappears until the end of the current review behavior segment, the consecutively fitted trajectory positions are not included in the score deviation continuation segment, and the score deviation direction fixation amount is recorded as a non-segment-crossing state; if a score deviation direction before the turning point of the score deviation reappears after a fit, the fitted trajectory position and the trajectory position where the original direction reappears are both included in the score deviation continuation segment; if the opposite direction appears after a fit, the score deviation continuation segment ends.

[0088] When a scoring deviation continuation segment exists, the number of trajectory positions continuously covered by the continuation segment after crossing the turning point of the scoring deviation is counted and used as the scoring deviation direction fixation quantity. When a scoring deviation continuation segment does not exist, the scoring deviation direction fixation quantity is recorded as an uncrossed segment state. The dimension of the scoring deviation direction fixation quantity is the number of trajectory positions, used to express the length of time the scoring deviation direction continues to remain in the original direction after the turning point of the scoring deviation.

[0089] In one embodiment, in an expert's review behavior trajectory, there are consecutive instances of scores being too high before the turning point of score deviation. The first trajectory position after the turning point is a "close" position, the next trajectory position shows another instance of being too high, and the next trajectory position turns to a "low" position. Since the "close" position is followed by another instance of being too high before the turning point, both the "close" and "high" positions are included in the score deviation continuation segment. The score deviation continuation segment ends when the subsequent trajectory position turns to a "low" position. The fixation amount of the score deviation direction is determined by the number of trajectory positions covered by the score deviation continuation segment.

[0090] After S302, whether the scoring deviation direction still maintains its original direction after the turning point of the scoring deviation is converted into the number of traceable continuous trajectory positions.

[0091] S303 Sequence Crossover Results Formation and Expert Behavioral Profile Determination.

[0092] S303 takes the review time transition lag obtained from S301 and the score deviation direction fixation obtained from S302 as inputs to form a sequence crossover result within the same review behavior trajectory. The sequence crossover result is the structural discrimination result in adjacent review behavior segments, used to determine whether the change in review time and the change in score deviation are misaligned near the transition segment of score deviation. The specific meaning of misalignment is that the review time transition position is already after the end boundary of the transition segment of score deviation, while the score deviation direction continues in its original direction after the transition segment of score deviation.

[0093] Before generating the ordinal crossover results, it is first confirmed that both the review time transition lag and the score deviation direction fixation correspond to the same score deviation transition segment. If the review time transition lag and the score deviation direction fixation correspond to different score deviation transition segments, then ordinal crossover analysis is performed under their respective score deviation transition segments, without performing cross-transition segment merging.

[0094] When the review time lag is equal to the number of trajectory positions, and the score deviation direction fixation is also equal to the number of trajectory positions, the sequence crossover result points to the conflict segment. The conflict segment covers the trajectory interval where the score deviation transition segment, the review time lag position, and the score deviation continuation segment are located. At this time, the time alignment recalculation only processes the historical review process covered by the conflict segment and the adjacent historical review processes on both sides of the conflict segment. The time alignment recalculation re-verifies the correspondence between the acceptance time, submission time, and trajectory pre-sequence according to the time alignment rules in step S1, and re-forms the trajectory pre-sequence within the conflict segment range. The time alignment recalculation does not change the expert index or the already completed identity merging results.

[0095] When the review time lag is within a segment, a preceding segment, or an unformed segment, or when the score deviation direction fixation is in a segment-independent or unformed segment, the sequence crossover result does not point to a conflicting segment. In this case, the review behavior segment following the score deviation lag segment is determined as the expert behavior profile. The expert behavior profile includes the trajectory position after the score deviation lag segment, the corresponding historical review process, the score deviation position, and the review time position. If there are multiple score deviation lag segments in the review behavior trajectory, S301 to S303 are executed sequentially according to the trajectory pre-sequence, and it is determined whether the corresponding adjacent review behavior segments form a conflicting segment.

[0096] When the input object is a review behavior segment without a turning point, the calculation of the fixed amount of scoring deviation direction is not performed. If the scoring deviation positions in the review behavior segment without a turning point are all aligned, and the review time position does not form a subsequent shift, then the review behavior segment without a turning point is identified as an expert behavior profile; if the review time position in the review behavior segment without a turning point forms a subsequent shift, then an expert behavior profile is not formed, and the review behavior segment without a turning point is retained as a review behavior segment to be verified.

[0097] After S303, the lag in the review time transition and the fixation of the scoring deviation direction are placed in the same review behavior trajectory for structural discrimination. The conflict segment trigger time is aligned and recalculated. The review behavior segment after the transition segment of the scoring deviation that has not formed a conflict segment forms an expert behavior profile.

[0098] Step S3 unifies the direction of review time consumption and the direction of score deviation into the review behavior trajectory for ordinal cross-analysis, and outputs conflict segments or expert behavior profiles, so that the expert behavior profiles come from review behavior segments that have been jointly tested by the score deviation boundary and the change of review time.

[0099] Detailed implementation of step S4:

[0100] Step S3 has created an expert behavior profile, which is used to express the status of the review behavior that was not excluded from the conflict segment in the expert's historical review behavior. However, the expert behavior profile cannot indicate whether the expert's academic information covers the research direction of the submitted paper data. Therefore, step S4 needs to first organize the submitted paper data and expert academic information into the same field, and then obtain the candidate expert set based on the main research direction and auxiliary direction.

[0101] The S401 submitted paper indicates the formation of the paper.

[0102] Step S401 takes the submitted paper data obtained in step S1 as input. The submitted paper data includes the paper title, research direction, and keywords. At the start of processing, the research direction field is read first. If the research direction field exists, the direction description in the research direction field is determined as the main research direction. If the research direction field is empty, words that can express the discipline scope and research object are read from the paper title, and the direction description that simultaneously points to both the discipline scope and research object is determined as the main research direction. When determining the main research direction from the paper title, the paper title is first matched against the direction name, discipline name, and research object name in the research direction directory of the blind review business system. When the paper title matches both the discipline name and the research object name, the research direction directory entry that corresponds to both is determined as the main research direction. When the paper title matches multiple research direction directory entries, the research direction directory entry that matches both the discipline name and the research object name is selected first. When the paper title cannot match any research direction directory entry, the submitted paper indicates that it does not form a main research direction, and the corresponding submitted paper data is not included in the candidate expert set for processing. Keywords are not used to determine the main research direction; they are only used to form auxiliary directions.

[0103] After the main research direction is determined, the processing continues by reading the paper title and keywords. The subject description in the paper title limits the disciplinary scope of the submitted paper data; the research object in the title limits the paper's focus; and the core description in the keywords supplements the research object and methodology. For repeated directional descriptions, only the first occurrence is retained; words that only indicate writing actions such as research, analysis, exploration, and application are not included in the submitted paper's description, as these words cannot distinguish the research direction. For expressions with the same meaning but different writing styles, they are organized according to the descriptions used in the research direction field of the blind review system.

[0104] When generating the representation of submitted papers, the processing steps store the main research direction, discipline description, research object, research method, and auxiliary directions into corresponding fields, and retain a source tag for each field. The source tag indicates that the field content comes from the paper title, research direction, or keywords. The source tag does not participate in sorting or change the field content; it is only used to distinguish between the main research direction and auxiliary directions in S403. After S401, the submitted paper data is organized into a representation of submitted papers with a clear field structure, which can express the scope of the paper's topic, main research direction, and auxiliary directions.

[0105] S402 Expert Academic Expression Formulation.

[0106] S402 takes expert academic information from the profile baseline record as input, including expert index, subject name, and research direction. The processing reads expert academic information one by one according to the expert index, and organizes the subject name and research direction into the same field format as the submitted paper. The subject name is written into the subject description field of the expert academic representation, and the research direction is written into the direction description field. If multiple research directions exist under the same expert index, they are retained in their original order in the profile baseline record and written into the direction description field respectively.

[0107] The expert academic statements follow the same organization rules as S401. For repeated expressions in the expert academic information, only the first occurrence is retained; general descriptions that do not express the scope of the discipline, research object, or research method are not included in the expert academic statements; for research directions with the same meaning but different writing styles, they are organized according to the commonly used expressions of discipline names and research directions in the basic profile record. If the discipline name and research direction in the expert academic information belong to different disciplines, both the discipline name and research direction are retained, and when entering S403, they are respectively compared with the submitted paper's representation for field matching; they are not merged or rewritten in S402.

[0108] After the expert academic representation is formed, each expert index corresponds to one expert academic representation. The expert academic representation is consistent with the expert index in the profile base record and retains the correspondence with the expert behavior profile formed in step S3. After step S402, the expert academic information is organized into expert academic representations that can be compared with the representation of the submitted paper using the same field caliber.

[0109] S403 candidate expert set obtained.

[0110] S403 takes the submitted paper representation generated in S401 and the expert academic representation generated in S402 as inputs, and places them into the same semantic representation space for similarity judgment. In this step, the semantic representation space refers to a unified field space composed of discipline description, direction description, research object, research method, and auxiliary direction. The submitted paper representation and the expert academic representation are compared under the same field meaning.

[0111] The similarity assessment follows the order of primary research direction first, followed by secondary research directions. First, the primary research direction stated in the submitted paper is read, then the subject and research direction descriptions in the expert's academic statement are read. If the subject or research direction description in the expert's academic statement covers the primary research direction stated in the submitted paper, then the secondary research direction assessment proceeds; if neither the subject nor research direction description in the expert's academic statement covers the primary research direction stated in the submitted paper, then the corresponding expert index is not included in the candidate expert set.

[0112] The determination of auxiliary research directions is performed after the main research direction has been approved. The process reads the research object, research method, and auxiliary research directions from the submitted paper, and also reads the direction descriptions from the expert's academic statement. If the direction description in the expert's academic statement corresponds to the research object or research method in the submitted paper, the corresponding expert index is included in the candidate expert set. If the expert's academic statement only corresponds to the keyword auxiliary descriptions in the submitted paper but does not cover the main research direction, the corresponding expert index is not included in the candidate expert set.

[0113] The coverage of the main research direction is determined in the following order: Coverage is defined as follows: if the direction description in the expert's academic statement is consistent with the main research direction stated in the submitted paper; if the direction description in the expert's academic statement is a subordinate direction of the main research direction in the research direction directory of the blind review system; and if the subject description in the expert's academic statement covers the subject area of ​​the main research direction, and the direction description in the expert's academic statement corresponds to the research object stated in the submitted paper. The correspondence between research object and research method is determined based on field consistency, hierarchical relationships in the research direction directory, and the synonym relationships preset by the blind review system. The research direction directory originates from the subject directory, professional directory, and research direction entries pre-maintained in the blind review system. Each research direction entry in the research direction directory includes a direction name, subject area, superior direction, subordinate direction, and synonym. Synonym relationships are established by the blind review system based on direction aliases maintained by the submitting institution, standard names in the subject directory, and commonly used expressions in historical submitted paper data. Field consistency means that two fields have the same expression after textual standardization. Hierarchical relationship means that one direction description is a subordinate direction of another direction description in the research direction directory.

[0114] In one embodiment, the research direction of the submitted thesis data is educational evaluation, the thesis title includes blind review of graduate theses, and the keywords include expert profile and data analysis. S401 identifies educational evaluation as the primary research direction and categorizes blind review of graduate theses, expert profile, and data analysis as secondary directions. The expert academic representation of expert index E001 includes educational evaluation and graduate education management, covering the primary research direction and corresponding to the research object of blind review of graduate theses; therefore, expert index E001 is included in the candidate expert set. The expert academic representation of expert index E002 only includes data analysis; although it corresponds to the keyword "data analysis," it does not cover the primary research direction of educational evaluation; therefore, expert index E002 is not included in the candidate expert set.

[0115] When the candidate expert set is formed, each candidate expert retains an expert index, expert academic representation, and the source of the expert behavior profile corresponding to the expert index. After step S403, the experts in the candidate expert set have been judged by their main research direction and auxiliary research direction, and the candidate expert set can directly enter the expert behavior profile mapping and candidate ranking processing in step S5.

[0116] Step S4 organizes the submitted paper data and expert academic information into the same semantic representation space for the submitted paper representation and expert academic representation, and uses the coverage of the main research direction as a prerequisite for entering the candidate expert set, and uses the auxiliary direction correspondence to limit the academic scope of the candidate expert set.

[0117] Detailed implementation of step S5:

[0118] Step S4 has formed a set of candidate experts, and the indexes of the experts in the set have been determined by the main research direction between the submitted papers and the academic statements of the experts. However, the set of candidate experts has not yet formed a ranking order for assignment, and the actual assignment results have not yet been returned to the profile base record to verify the expert behavior profile. Therefore, step S5 needs to complete the formation of candidate records, the generation of candidate ranking results, and the feedback of actual assignment results.

[0119] S501 candidate records are formed.

[0120] S501 takes the candidate expert set formed in step S4 as input and uses the expert index as the sole connection basis. The process reads the expert index in the candidate expert set one by one, and then reads the expert's academic representation and expert behavioral profile corresponding to the same expert index from the profile base record. The expert's academic representation comes from step S4, and the expert behavioral profile comes from step S3. The two cannot be re-matched by expert name, institution, or discipline name to avoid mismatches between experts with the same name or experts from different institutions during the candidate stage.

[0121] When the same expert index contains both expert academic representation and expert behavioral profile, and the expert behavioral profile originates from a review behavior segment that does not point to a conflicting segment in the ordinal cross-validation result, the expert index, expert academic representation, and expert behavioral profile are combined into a candidate record. The expert academic representation in the candidate record maintains the field structure of step S4, including subject description, direction description, research object, research method, and auxiliary direction. The expert behavioral profile in the candidate record maintains the trajectory structure of step S3, including the trajectory position after the turning point of the scoring deviation, the scoring deviation position, the review time position, and the corresponding historical review process.

[0122] If no expert behavior profile exists in the same expert index, it means that step S3 failed to form a usable review behavior segment after the turning point of scoring deviation, and the corresponding expert index does not form a candidate record. If multiple expert behavior profiles exist in the same expert index, first read the trajectory position covered by each expert behavior profile, and then select the expert behavior profile with the latest trajectory preorder position. If the trajectory positions covered by multiple expert behavior profiles overlap, and the overlapping position coincides with the conflict segment, the corresponding expert index does not form a candidate record; if the overlapping position does not coincide with the conflict segment, select the expert behavior profile covering the latest trajectory position. After the above processing, the candidate record has both academic direction source and review behavior source, and both can return to the same expert index.

[0123] The candidate ranking results for S502 and the list of blind review experts were formed.

[0124] S502 takes the candidate records formed in S501 as the processing object, first reads the main research direction, research object, research method and auxiliary direction in the submitted paper, and then reads the academic statements of experts in the candidate records. Since the candidate expert set has been judged by the main research direction in step S4, this sub-step no longer judges whether the candidate records cover the main research direction, but compares the degree of correspondence between the candidate records and the research object and research method under the same main research direction.

[0125] In specific processing, when the directional description in the expert's academic statement corresponds to both the research object and research method in the submitted paper, the corresponding candidate record is ranked before candidate records that only correspond to the research object or only correspond to the research method. When the directional descriptions in the expert's academic statement only correspond to auxiliary directions, the expert behavioral profile is compared further. The closer the trajectory position covered by the expert behavioral profile is to the end of the review behavior trajectory, the better it reflects the expert's recent review behavior status. Therefore, among candidate records with the same degree of academic correspondence, the candidate record whose expert behavioral profile covers the latest trajectory position is prioritized. If both the degree of academic correspondence and the latest trajectory position of the expert behavioral profile are the same, they are arranged according to the fixed order of the expert index in the blind review business system to ensure that the candidate ranking results can be reproduced.

[0126] When a submitted paper does not specify a research method, candidate records are ranked solely based on the degree of correspondence with the research object. When a submitted paper does not specify a research object but does specify a research method, candidate records are ranked solely based on the degree of correspondence with the research method. When neither a research object nor a research method is specified in the submitted paper, candidate records are ranked according to the degree of correspondence with the auxiliary direction, and then ranked according to the latest trajectory position covered by the expert behavior profile. If the number of expert indexes in the candidate ranking results is less than the required number of reviewers, a blind review expert list is formed based on the existing candidate ranking results, and a supplementary reviewer marker is retained in the profile base record. The supplementary reviewer marker indicates that the current number of candidate ranking results is insufficient and does not change the already formed candidate ranking results.

[0127] When generating the candidate ranking results, the source of the expert behavior profile in each candidate record is read. If the expert behavior profile originates from a review behavior segment whose order cross-result does not point to a conflict segment, the candidate record is included in the candidate ranking results; if the expert behavior profile is missing, or if the expert behavior profile originates from a conflict segment, the candidate record is not included in the candidate ranking results. After the candidate ranking results are generated, the expert index is obtained sequentially from the candidate ranking results according to the number of papers submitted for review required by the blind review business system, forming a blind review expert list. The blind review expert list retains the expert index, expert academic representation, source of expert behavior profile, and arrangement position in the candidate ranking results, without changing the submitted paper representation and expert academic representation already generated in step S4.

[0128] After S502, the candidate expert set is organized into a candidate ranking result with a clear order. The experts in the blind review expert list not only indicate that they are in the same main research direction as the submitted papers, but also have a clear source of expert behavior profiles.

[0129] S503 Actual Interrogation Results Feedback and Portrait Base Record Update.

[0130] S503 is executed after the blind review expert list is completed and the actual review results corresponding to the blind review expert list are retrieved. The actual review results include the expert index, the data identifier of the submitted paper, the review acceptance time, the submission time, the submission deadline, and the scoring result. The processing first locates the basic profile record based on the expert index, and then incorporates the actual review results as a new historical review process into the basic profile record under the corresponding expert index. If the review acceptance time exists, the actual review result is placed in the review behavior trajectory according to the review acceptance time; if the review acceptance time is missing but the submission time exists, the actual review result is placed in the review behavior trajectory according to the submission time; if both the review acceptance time and the submission time are missing, the actual review result is not included in the review behavior trajectory and is retained as an unsorted record.

[0131] After the actual review results enter the review behavior trajectory, the scoring deviation position is determined according to the scoring deviation position determination rules in step S2. Specifically, firstly, the set of review results corresponding to the same submitted paper is obtained based on the data identifier of the submitted paper; if there are no comparable review results for the same submitted paper, the set of review results from the same batch and the same research direction is obtained; then, the scoring results in the actual review results are compared with the corresponding positions in the review result set to obtain the scoring deviation positions that are too high, too low, or too close. At the same time, according to the review time position determination rules in step S3, the review time position of the actual review results is determined by using the acceptance time, submission time, and review cycle of the submission task.

[0132] After determining the scoring deviation position and the review time position, the meaning of interrupting the scoring deviation direction in the expert behavior profile is first defined. Interrupting the scoring deviation direction in the expert behavior profile means that the effective scoring deviation position closest to the end of the review behavior trajectory in the expert behavior profile is too high, while the scoring deviation position at the actual review result's trajectory position is too low; or the effective scoring deviation position closest to the end of the review behavior trajectory in the expert behavior profile is too low, while the scoring deviation position at the actual review result's trajectory position is too high. If the scoring deviation position at the actual review result's trajectory position is a close match or an incomparable scoring position, it is not considered an interruption of the scoring deviation direction in the expert behavior profile.

[0133] A partial verification is conducted based on the trajectory location of the actual review result. The scope of the partial verification includes the trajectory location of the actual review result, the preceding trajectory location, and the following trajectory location; if the following trajectory location does not yet exist, the scope of the partial verification includes the trajectory location of the actual review result and the preceding trajectory location. If the following trajectory location does not yet exist, and the review time position of the trajectory location of the actual review result is shifted relative to the preceding trajectory location, then the trajectory location of the actual review result is recorded as a turning point in review time to be confirmed, without immediately forming a new conflict segment. Once the next historical review process enters the review behavior trajectory of the same expert index, a partial verification is then performed based on the trajectory location of the actual review result, the following trajectory location, and the scoring deviation position.

[0134] The partial review first determines whether a new turning point in scoring deviation has formed within the scope of the partial review, according to the rules for determining turning points in scoring deviation. If no new turning point in scoring deviation has formed within the scope of the partial review, the actual assignment result will not form a new conflict segment in the review behavior trajectory. If a new turning point in scoring deviation has formed within the scope of the partial review, the new review time turning point lag and the new scoring deviation direction fixation are calculated, and a new ordinal cross result is formed. When the new ordinal cross result points to a conflict segment, the actual assignment result forms a new conflict segment in the review behavior trajectory; when the new ordinal cross result does not point to a conflict segment, the expert behavior profile is maintained. When a new conflict segment is formed, the boundary of the turning point in scoring deviation is redefined based on the scope of the partial review. After the boundary of the turning point in scoring deviation is updated, the review behavior segment after the turning point in scoring deviation is redefined, and the redefined review behavior segment is used as the expert behavior profile. The update of the boundary of the turning point in scoring deviation only affects the review behavior trajectory in the profile's basic record and does not change the expert index, expert basic attributes, expert academic information, or expert academic representation.

[0135] In one embodiment, expert index E001 has been included in the blind review expert list, and after the actual review results are generated, it is inserted at the end of E001's review behavior trajectory. If the newly inserted trajectory position is relatively low compared to the set of review results in the same batch and research direction, and the effective score deviation position closest to the end of the review behavior trajectory in E001's original expert behavior profile is also relatively low, and the review time position of the newly inserted trajectory position has not shifted backward compared to the previous trajectory position, then E001's expert behavior profile is maintained. If the newly inserted trajectory position changes from the original relatively low direction to relatively high, and the review time position shifts backward compared to the previous trajectory position, but there is no subsequent trajectory position at the end of the review behavior trajectory, then the newly inserted trajectory position is recorded as the review time turning point to be confirmed; after the next historical review process enters the review behavior trajectory, if the backward state continues to occur, and a new turning point of score deviation is formed within the local verification scope, then it is determined whether a new conflict segment has been formed based on the new order cross result.

[0136] After S503, the actual review results are reflected back to the profile base record. The new scoring deviation position and review time position can be partially verified in the original review behavior trajectory. The expert behavior profile is maintained or updated according to the formation of new conflict segments.

[0137] Step S5 converts the candidate expert set into candidate ranking results and a blind review expert list, and reflects the actual review results back to the profile base record, so that the boundary of the turning point of the score deviation can be locally updated according to the new blind review process.

[0138] The above embodiments are merely preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for constructing expert profiles for blind review of graduate theses, characterized in that, include: S1. Obtain the submitted paper data and expert historical data from the blind review business system, merge the expert historical data by identity, unify the format and align the time, and form a basic record of profiles with experts as the index. S2. Based on the time alignment results in the basic portrait record, arrange the historical review process of each expert into a review behavior trajectory. Determine the turning point of the score deviation according to the position of the score deviation in the review behavior trajectory, and use the turning point of the score deviation as the boundary to form the review behavior segment. S3. Calculate the review time transition lag and the score deviation direction fixation for adjacent review behavior segments. Based on the review time transition lag and the score deviation direction fixation, form the ordinal cross result. When the ordinal cross result points to the conflict segment, recalculate the time alignment of the profile basic record. Otherwise, determine the review behavior segment after the score deviation transition segment as the expert behavior profile. The review time transition lag is formed as follows: when both the acceptance time, submission time, and submission deadline exist, and the submission deadline is later than the acceptance time, the progress position of the submission time in the review cycle of the submission task formed by the acceptance time and the submission deadline is determined as the review time position; if the acceptance time, submission time, or submission deadline is missing, the corresponding trajectory position will not form a review time position; when the review time position changes from moving forward or remaining flat to moving backward and the backward movement continues to appear in the next adjacent trajectory position, a review time transition position is formed, and the order cross result is formed only when the review time transition lag and the score deviation direction fixation amount correspond to the same score deviation transition segment; S4. Perform semantic representation processing on the submitted paper data and the expert academic information in the profile basic record to form the submitted paper representation and the expert academic representation. Obtain the candidate expert set based on the similarity between the submitted paper representation and the expert academic representation. S5. Map the expert behavior profiles in the candidate expert set to the corresponding expert academic representations to form the candidate ranking results. Based on the candidate ranking results, form a blind review expert list and reflect the actual review results corresponding to the blind review expert list back to the profile base record. When the actual review results form a new conflict segment in the review behavior trajectory, update the boundary of the turning point of the score deviation; otherwise, maintain the expert behavior profile.

2. The method for constructing expert profiles for blind review scenarios of graduate theses according to claim 1, characterized in that, Identity merging includes: when expert identifiers exist and are consistent, the corresponding historical review processes are grouped into the same expert index; when at least one historical review process lacks an expert identifier, the expert name, institution, and discipline name are first processed before merging, and then, when the expert name, institution, and discipline name are all consistent, they are grouped into the same expert index; when any field is inconsistent, a record to be merged is formed; the format is standardized by organizing the historical review process into fields such as submitted paper data identifier, review time, submission time, submission deadline, and scoring result, and the completeness of the record is formed based on the existence of the fields.

3. The expert profile construction method for blind review scenarios of graduate theses according to claim 1, characterized in that, The scoring deviation position is determined as follows: Based on the data identifier of the submitted paper, obtain the review result set corresponding to the same submitted paper; when the review result set contains scoring results other than the current trajectory position, perform directional comparison; when the review result set lacks comparable review results, obtain the review result set of the same batch and the same research direction; when neither of the two types of review result sets is sufficient to form a control position, record the current trajectory position as an incomparable scoring position; when a control position is obtained, record the scoring result as higher, lower, or close to the control position based on its height or falling relationship with the control position, and form a turning point of scoring deviation when the higher or lower position changes direction and the changed direction continues in subsequent adjacent trajectory positions.

4. The method for constructing expert profiles for blind review scenarios of graduate theses according to claim 1, characterized in that, The main research direction in the submitted paper is determined as follows: When the research direction field exists, the direction description in the research direction field is determined as the main research direction; when the research direction field is empty, the paper title is matched with the direction name, discipline name, and research object name in the research direction directory. If the paper title matches an entry in the research direction directory, the main research direction is determined. If the paper title does not match an entry in the research direction directory, the submitted paper does not form a main research direction and is not included in the candidate expert set for processing; if the direction description in the expert academic statement is consistent with the main research direction and belongs to a subordinate direction of the main research direction, or if the discipline description in the expert academic statement covers the discipline scope of the main research direction and the direction description corresponds to the research object, it is determined to cover the main research direction.

5. The method for constructing expert profiles for blind review scenarios of graduate theses according to claim 1, characterized in that, When there are incomparable scoring positions in the review behavior trajectory, these incomparable scoring positions are not included in the confirmation of the turning point of scoring deviation and are retained in the review behavior trajectory. When there are no turning points of scoring deviation in the review behavior trajectory, the entire review behavior trajectory forms a no-turn review behavior segment. The no-turn review behavior segment is not divided into the previous review behavior segment and the next review behavior segment, and is processed in the no-turn review behavior segment. When all scoring deviation positions in the no-turn review behavior segment are aligned and the review time position does not form a backward continuation, the no-turn review behavior segment is identified as the expert behavior profile. When the review time in a review behavior segment without a turning point is shifted and continued, no expert behavior profile is formed, and it is retained as a review behavior segment pending verification.

6. The method for constructing expert profiles for blind review scenarios of graduate theses according to claim 1, characterized in that, When forming the blind review expert list, if the number of expert indexes in the candidate ranking results is less than the required number of reviewers, the blind review expert list is formed according to the existing candidate ranking results, and a supplementary reviewer mark is retained in the profile basic record. When the actual review result is inserted at the end of the review behavior trajectory and there is no subsequent trajectory position, if the review time position of the actual review result is shifted to the previous trajectory position, the trajectory position of the actual review result is recorded as the turning point of review time to be confirmed and no new conflict segment is formed; after the next historical review process enters the review behavior trajectory of the same expert index, the local verification continues to be performed based on the trajectory position of the actual review result, the subsequent trajectory position, and the scoring deviation position.

7. The method for constructing expert profiles for blind review scenarios of graduate theses according to claim 2, characterized in that, The profile's basic records include the expert's basic attributes, academic information, historical review processes with pre-ordered trajectories, records to be merged, and verification records under the same expert index. Historical review processes with pre-ordered trajectories retain the data identifier of the submitted paper, the time of acceptance, the time of submission, the deadline for submission, and the scoring results. Records to be merged retain the original expert's name, institution, and discipline. Verification records retain duplicate historical review processes that did not participate in time alignment. When both the time of acceptance and the time of submission are missing, the corresponding historical review process is retained as an unsorted record.

8. The method for constructing expert profiles for blind review scenarios of graduate theses according to claim 4, characterized in that, Each candidate expert in the candidate expert set retains an expert index, expert academic representation, and an entry point for the expert behavior profile corresponding to the expert index; Each candidate record in the candidate ranking results retains the expert index, expert academic representation, source of expert behavioral profile, and ranking position; the blind review expert list obtains the expert index according to the ranking position of the candidate ranking results, and retains the expert academic representation, source of expert behavioral profile, and ranking position in the candidate ranking results; when the number of expert indexes in the candidate ranking results is less than the required number of review assignments, a supplementary review assignment mark is retained in the profile base record.

9. The method for constructing expert profiles for blind review scenarios of graduate theses according to claim 6, characterized in that, The actual review results include expert index, submitted paper data identifier, review time, submission time, submission deadline, and scoring results. After the actual review results are reflected in the basic profile record, the landing point of the actual review results in the review behavior trajectory is determined according to the review time or submission time. If both the review time and submission time are missing, the actual review results are retained as unsorted records. After the location of the actual review results is recorded as the turning point of the review time to be confirmed, when the next historical review process enters the review behavior trajectory and forms a backward continuation, it is first determined whether a new turning point of scoring deviation has been formed within the local verification scope. Then, a new ordinal cross result is formed based on the new review time turning point lag and the new scoring deviation direction fixation. When the new ordinal cross result points to the conflict segment, the boundary of the turning point of scoring deviation is updated.

Images