Method and apparatus for recognizing new words and first occurrence events

Abstract

The application discloses a new word and a first occurrence event recognition method and device, wherein the method comprises the following steps: obtaining hot-line work order text data according to a preset period, and preprocessing the hot-line work order text data to obtain a structured text stream; performing word segmentation and sliding window processing on the text stream to obtain an initial phrase candidate set; performing statistical access screening on the initial phrase candidate set, and performing alias merging on the candidate phrases passing the statistical access by using consistency rules of specified dimensions to obtain a word candidate set; updating a current space-time reference baseline based on the word candidate set, and recalculating historical full-amount data according to a specified fixed period to re-calibrate the space-time reference baseline; and determining whether there is a new word in the word candidate set and whether there is a first occurrence event based on the re-calibrated space-time reference baseline. Based on the unified space-time reference baseline, the new word and the first occurrence can be accurately and timely determined.

Description

Technical Field

[0001] This invention relates to the field of information processing technology, specifically to a method and apparatus for identifying new words and first occurrence events. Background Technology

[0002] In related technologies, large-scale, time-sensitive, and conversational call texts are generated daily through the Beijing 12345 hotline or other information receiving channels. To detect "new, first-time, and sudden" events as early as possible, existing services generally use "new word discovery" as an early signal. This involves identifying segments not included in the historical word list or with significantly abnormal statistical distribution within a continuous time window, i.e., "new words," and using these to form a list of suspected event clues for review or alert.

[0003] The above-mentioned scheme, which uses "new word discovery" as the core of the first-time event identification technology, can serve as an early warning method, but it has inherent shortcomings in semantic alignment, spatiotemporal consistency, cross-document evidence merging and interpretability, making it difficult to reliably support accurate and timely identification at the "new / first-time" level. Summary of the Invention

[0004] The main objective of this invention is to provide a method and apparatus for identifying new words and first-time events, so as to overcome the shortcomings of related technologies.

[0005] To achieve the above objectives, according to a first aspect of the present invention, a method for identifying new words and first-time events is provided, comprising: acquiring hotline work order text data at a preset period, and preprocessing the hotline work order text data to obtain a structured text stream; performing word segmentation and sliding window processing on the text stream to obtain an initial phrase candidate set; performing statistical admission screening on the initial phrase candidate set, and merging the candidate phrases that pass the statistical admission using a consistency rule of a specified dimension to obtain a term candidate set; updating the current spatiotemporal reference baseline based on the term candidate set, and recalculating the entire historical data at a specified fixed period to recalibrate the spatiotemporal reference baseline; and determining whether there are new words and first-time events in the term candidate set based on the recalibrated spatiotemporal reference baseline.

[0006] According to a second aspect of the present invention, a device for identifying new words and first-time events is provided, comprising: a preprocessing unit, configured to acquire hotline work order text data at a preset period and preprocess the hotline work order text data to obtain a structured text stream; a term candidate set determination unit, configured to perform word segmentation and sliding window processing on the text stream to obtain an initial phrase candidate set; perform statistical admission screening on the initial phrase candidate set, and perform alias merging on the candidate phrases that pass the statistical admission using a consistency rule of a specified dimension to obtain a term candidate set; a calibration unit, configured to update the current spatiotemporal reference baseline based on the term candidate set, and recalculate the entire historical data at a specified fixed period to recalibrate the spatiotemporal reference baseline; and a new word and first-time event determination unit, configured to determine whether there are new words in the term candidate set and whether there are first-time events based on the recalibrated spatiotemporal reference baseline.

[0007] This embodiment of the device for identifying new words and first-time events includes a method comprising: acquiring hotline work order text data at a preset period, preprocessing the hotline work order text data to obtain a structured text stream; performing word segmentation and sliding window processing on the text stream to obtain an initial phrase candidate set; performing statistical admission screening on the initial phrase candidate set, and merging the candidate phrases that pass the statistical admission using a consistency rule of a specified dimension to obtain a term candidate set; updating the current spatiotemporal reference baseline based on the term candidate set, and recalculating the entire historical data at a specified fixed period to recalibrate the spatiotemporal reference baseline; and determining whether there are new words and first-time events in the term candidate set based on the recalibrated spatiotemporal reference baseline. By addressing the high concurrency, timeliness, and colloquial nature of hotline text streams, an integrated new word identification and first-time event determination system is constructed, with "unified hierarchical baseline, robust candidates, evidence-based decision-making, persistent trends, risk quantification, and adaptive threshold control" as its core principles. The system retains the high sensitivity of statistical methods while ensuring that "new words can reliably represent new events" through business semantics and evidence chains. It can output stable judgments in the first to the first three incoming calls and maintain consistent standards and auditability across multiple districts and time periods throughout the city, thereby improving accuracy and timeliness. Attached Figure Description

[0008] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0009] Figure 1 This is a flowchart of the method for identifying new words and first-time events according to an embodiment of the present invention;

[0010] Figure 2 This is a schematic diagram of an electronic device according to an embodiment of the present invention. Detailed Implementation

[0011] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. 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 should fall within the scope of protection of the present invention.

[0012] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate for the embodiments of the invention described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0013] It should be noted that, unless otherwise specified, the embodiments and features described in the present invention can be combined with each other. The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0014] According to embodiments of the present invention, a method for identifying new words and first occurrence events is provided, such as... Figure 1 As shown, steps 101 to 103 are included below:

[0015] Step 101: Acquire hotline work order text data according to a preset cycle, and preprocess the hotline work order text data to obtain a structured text stream.

[0016] In this step, the work order text and metadata of the 12345 hotline are accessed by minute or hour. The metadata includes at least timestamps, zoning information, channel identifiers, work order numbers, topic tags, and calendar identifiers for holidays and major events. The work order text is then processed sequentially with desensitization, sentence segmentation, colloquial synonymization, standardization of number and time formats, two-layer mapping of addresses and points of interest, and text denoising and error correction, outputting a structured and consistent text stream.

[0017] Step 102: Perform word segmentation and sliding window processing on the text stream to obtain an initial phrase candidate set; perform statistical admission screening on the initial phrase candidate set, and merge the candidate phrases that pass the statistical admission using a consistency rule of a specified dimension to obtain a term candidate set.

[0018] In this step, based on the consistent text stream output, word segmentation and word-level n-gram sliding window (e.g., n takes the value of 2-3) are used in parallel to generate an initial phrase candidate set. Statistical admission screening is performed on the initial phrase candidate set, with the following criteria: mutual information value greater than a preset mutual information threshold, log-likelihood ratio greater than a preset log-likelihood ratio threshold, left and right entropy values ​​greater than preset left and right entropy thresholds, and frequency of occurrence greater than a preset minimum frequency. For candidate phrases that pass the statistical admission, a three-channel consistency rule of "homophonic similarity - semantic vector - left and right entropy" is used to merge aliases, resulting in term-level candidates. The candidate segments generated by word segmentation and word-level sliding window are combined with statistical admission criteria such as mutual information, left and right entropy, and log-likelihood, and aliases and misspellings are merged using a three-channel consistency rule of homophonic similarity, edit distance, and vector similarity to unify "multiple names for the same thing".

[0019] For example, the criteria for determining the consistency of the three channels can be homophonous similarity > 0.8, semantic vector cosine similarity > 0.7, and left and right entropy difference < 0.3, and merging is performed only when all three criteria are met simultaneously.

[0020] Step 103: Update the current spatiotemporal reference baseline based on the term candidate set, and recalculate the entire historical data at a specified fixed period to recalibrate the spatiotemporal reference baseline; determine whether there are new words in the term candidate set and whether there is a first occurrence event based on the recalibrated spatiotemporal reference baseline.

[0021] In this step, the spatiotemporal reference baseline is a historical data benchmark used to integrate the temporal and spatial dimensions. Its purpose is to quantify the degree of deviation between the probability of the current event and historical patterns, thereby supporting the accurate identification of abnormal events in dynamic spatiotemporal data. This baseline can eliminate random noise by establishing a unified spatiotemporal statistical reference system, maintain statistical consistency across regions and time periods, prevent weak data areas from being masked by strong data areas, and provide stable and comparable reference data for new word determination and first event identification.

[0022] The spatiotemporal reference baseline constructs a spatiotemporal joint probability distribution model based on historical data, specifically including the following technical elements:

[0023] 1. Time dimension: Layered time windows (including day, hour, and minute levels) are used to capture periodic patterns, and seasonalization is achieved by separating holiday rhythms and diurnal rhythms, thereby reducing the interference of periodic changes on anomaly detection;

[0024] 2. Spatial dimension: Geographical regions are divided according to administrative districts, street grids, etc., to characterize the distribution of spatially clustered events;

[0025] 3. Weak region correction mechanism: For regions with sparse sample sizes, the hierarchical Bayesian shrinkage method is used to adjust the statistical results to the prior of the higher-level regions in order to improve the statistical stability in the case of small samples.

[0026] 4. Dynamic update mode: Combining the online minute-level sliding window incremental update method with the offline hour-level or day-level batch recalculation baseline calibration method, a hybrid update of the spatiotemporal reference baseline is achieved to ensure that the detection results have both the timeliness of event response and maintain long-term statistical accuracy.

[0027] 5. Output indicators: Under a unified statistical standard, output the expected probability and confidence interval of each term in a specific region and time period. This indicator can be used as a quantitative basis for determining whether a term deviates from historical patterns.

[0028] Through the above-mentioned hierarchical modeling combined with real-time driven dynamic updates, the spatiotemporal reference baseline can adapt to complex spatiotemporal change patterns and provide stable and auditable abnormal event judgment results in the early stage of the first occurrence of an event (such as the first to the first three incoming calls).

[0029] When dynamically updating the base, this process must be executed in each round of new word discovery, employing a hybrid mode of online incremental updates and offline batch recalculation to balance timeliness and statistical stability. For example, during online processing, minute-level sliding window incremental updates target real-time access work order data, using minute-level sliding window granularity (e.g., 5 minutes, 15 minutes). For candidate terms within the current window, real-time incremental updates are performed on their statistics under the corresponding "district-time period" dimension (e.g., the frequency of the term "leakage" under the "Chaoyang District - 14:00-14:05" dimension within the current window). This ensures rapid response to newly generated data and supports time-sensitive scenarios such as rapid determination of the first incoming call.

[0030] When processing offline, recalculation is performed on all historical data (including historical cumulative data and real-time data updated incrementally on the online end) in batches at fixed intervals of "hourly" (e.g., every 1 hour) or "daily" (e.g., every early morning). This recalibrates the statistical baseline of all candidate terms under the "region-time period" dimension, eliminates short-term noise interference that may exist in the incremental updates on the online end, and ensures the long-term statistical accuracy of the baseline.

[0031] After obtaining the calibration baseline, time series decomposition is used to extract residuals to achieve deseasoning. For weak regions (sparse sample areas), the system applies hierarchical Bayesian shrinkage to shrink back to the city-level prior data for stable estimation. Finally, the system outputs a unified expected probability (the expected proportion of a term appearing in a specific time period) and a fluctuation range (confidence interval, representing the normal fluctuation range). By unifying the baseline, regional and rhythmic biases are eliminated, providing a consistent reference for subsequent new word determination.

[0032] As an optional implementation of this embodiment, determining whether there are new words in the candidate word set based on the recalibrated spatiotemporal reference baseline includes: outputting the expected probability and fluctuation range of the word in a unified manner according to time and space based on the recalibrated spatiotemporal reference baseline; calculating a preset index value based on the expected probability and fluctuation range of the word; calculating a new word index based on the preset index value; calculating a small sample confidence score using the Wilson interval estimation method with Bayesian prior for the candidate word set, and determining the trend persistence of the word using an exponentially weighted moving average algorithm with hysteresis gate; and determining the words in the candidate word set that belong to new words based on the new word index, the small sample confidence score, and the trend persistence.

[0033] In this optional implementation, four components are integrated: cohesion (reflecting the internal cohesion of a term), diversity (reflecting the richness of the term's context), relative improvement (reflecting the difference between the term's current and historical baselines), and time freshness (reflecting the term's recent activity) to obtain a new word index that characterizes the newness of a term. For example, ,in

[0034] Cohesion (Coh) measures the cohesion within a word segment by combining mutual information or log-likelihood ratio with left and right entropy. It's a strength indicator of whether words within a segment are naturally formed, often derived from a combination of mutual information or log-likelihood and left and right entropy, used to eliminate accidental word combinations. Diversity (Div) is a normalized count of the number of different words adjacent to each other, suppressing templated accidental combinations. Lift is defined using a uniform baseline as a reference. The "Fresh" index uses a combination of near-window price increase (s) and time decay term for normalization, emphasizing "the closer, the newer." This new word index integrates structural, contextual, and temporal information on the same baseline, making the "newness of words" closer to the "newness of events."

[0035] For small sample confidence, a Wilson interval estimation method with Bayesian prior is used to stabilize the probability of term occurrence in the small sample scenario corresponding to the first three work orders, outputting a confidence metric value in the 0-1 interval to support very early judgment. In the small sample stage, interval estimation with prior is often used to measure the reliability of new occurrence judgments, avoiding being misled by one or two accidental samples.

[0036] For example, a Bayesian prior-driven Wilson interval is used for early small samples: Where x is the number of work orders related to the term, and n is the total number of work orders in the current window. For the stratified baseline expectation, It represents the standard normal distribution function. This index still provides stable confidence even with a small sample size.

[0037] For trend persistence, an Exponentially Weighted Moving Average (EWMA) algorithm with hysteresis is used to analyze the time series of the new word index. A sustained upward trend is defined only when the EWMA values ​​are positive for m consecutive windows (m being 3-5 windows), outputting a persistence metric in the 0-1 range to distinguish between one-off peaks and sustained events. The persistence score obtained by exponentially weighting the new word index time series and setting a hysteresis gate is used to differentiate between one-off spikes and continuous upward trends.

[0038] For example, an exponentially weighted moving average is used with a hysteresis gate set, and a temperature rise is determined only when the temperature remains positive for a certain number of consecutive windows; short spikes do not enter a sustained state. The persistence value, Persist, is jointly mapped to the interval between zero and one by the continuous positive duration and the near-window rising slope.

[0039] If a term-level candidate meets the following conditions: "new word index > preset new word threshold (value 0.6-0.8), small sample confidence > 0.7, and trend persistence > 0.6", it is determined to be a new word.

[0040] As an optional implementation of this embodiment, when determining whether a first event exists, the method includes: performing time series decomposition on the recalibrated spatiotemporal reference baseline to extract and remove holiday rhythms and diurnal rhythm components to obtain a residual baseline; selecting from the determined new words the words that have never appeared in the historical residual baseline and appear for the first time in the current window, and adding the words to the first event candidate pool; triggering a first event alarm when the first event candidate words in the first event candidate pool meet the set conditions.

[0041] In this optional implementation, the seasonalized residuals, obtained by decomposing regular components such as holidays, weekends, and diurnal rhythms on the baseline, are used for anomaly and new occurrence identification, avoiding systematic false alarms caused by rhythms. The hierarchical spatiotemporal baseline maintains historical means and variances according to regions and time periods, and a comparison system obtained after removing holiday and diurnal rhythms serves as a unified reference for all indicators. Its purpose is to eliminate the masking of weak regions and rhythmic disturbances by strong regions, making cross-regional and cross-time period comparability possible.

[0042] Rhythm decomposition was performed on the hierarchical spatiotemporal baseline to separate holiday rhythms and diurnal rhythms, and residuals were extracted to form residual baselines. For weak regions with fewer than 10 work orders, hierarchical Bayesian shrinkage was used to calibrate the statistics. Where c is the number of times the term appears in the current window, and N is the current window's exposure. The expected proportion under the stratified baseline, Take one to five percent of the total. This indicator is an estimated proportion of the first three stages.

[0043] If a term-level candidate has never appeared in the residual baseline and appears for the first time in the current window, it is included in the initial candidate pool. If the initial candidate meets the preset conditions, there is no need to wait for sample accumulation to trigger an early alarm.

[0044] As an optional implementation of this embodiment, triggering a first-launch alarm when the first-launch candidate terms in the first-launch candidate pool meet the set conditions includes: calculating the significance deviation value of the term, wherein the significance deviation value is used to represent the degree of standardized deviation of the term's occurrence probability from the historical baseline: , and For stratified historical mean and standard deviation, As a smoothing factor; calculate the short-term increase value of the term, where the short-term increase value represents the increase in the probability of the term appearing in the current window relative to the average of the last k windows: , The value is the smoothed average of the recent windows; if the significant deviation meets the first threshold and the short-term increase meets the second threshold, an initial warning is triggered.

[0045] In this optional implementation, the smoothing ratio is used to represent the logarithmic increase of the mean of several windows, which is used to characterize whether there has been a recent resurgence of the term and whether old words are being used again.

[0046] Short-term gains , The mean of several windows is used to characterize the re-use of old words and short-term surges.

[0047] Significant deviation , and For stratified historical mean and standard deviation, It is a smoothing factor used to determine whether the current level is significantly higher than the historical level.

[0048] For example, an early initial stock alert can be triggered if the significance of the initial stock candidate is greater than 1.8-2.2 and the short-term price increase is greater than 0.5-0.8.

[0049] As an optional implementation method in this embodiment, when the initial candidate terms in the initial candidate pool meet the set conditions, the initial alarm is triggered as follows: the risk level of the term is determined based on the preset risk term list, and a continuous score of 0-1 is output as the risk weight; if the risk weight hits the risk gate control with the risk weight greater than the third threshold, the initial alarm is triggered.

[0050] In this optional implementation, a continuous score from 0 to 1 is output based on a risk vocabulary and a lightweight semantic classifier, with higher weights for public safety, public interest, and sensitive public opinion categories. For example, an early alert is triggered when the risk weight of the initial candidate is greater than 0.8. The risk weight quantifies the urgency of topics such as public safety and public opinion from zero to one, serving as the "urgent" dimension in priority ranking.

[0051] As an optional implementation of this embodiment, the method further includes: calculating a comprehensive score for the first-time event based on P = N×Conf×Persist×R×E, where N is the new word index, Conf is the small sample confidence level, Persist is the trend persistence, R is the risk weight, and E is the evidence level; and ranking the new words that triggered the first-time event based on the comprehensive score.

[0052] In this optional implementation, a unified ranking score is obtained by multiplying the new word index, confidence level, trend persistence, risk weight, and evidence level. Higher scores indicate greater priority for analysis and action. (Comprehensive scoring) R represents the risk weight, and E represents the degree of evidence. A higher score indicates that the information is both novel and credible, the trend is sustainable, and the risks and evidence are sufficient. An alarm is triggered when the threshold is reached.

[0053] As an optional implementation method in this embodiment, the original work order data is traced back for the sorted new words; information of a specified dimension is extracted from the original work order data; the degree of evidence is determined based on the information of the specified dimension and an evidence package is output.

[0054] In this optional implementation, high-scoring new word sets are aggregated based on time (work order timestamp), space (regional location), and co-occurring words (semantically related words). Triggering new words (e.g., "power outage") are traced back to the original work order, and four elements—time, location, subject, and phenomenon—are extracted (e.g., time "2023-08-01", location "Chaoyang District", subject "power grid company", phenomenon "voltage instability"). Merging is performed based on aggregation thresholds for element coverage (the proportion of elements present), temporal proximity (time difference < Δt, e.g., 1 hour; spatial distance < Δd, e.g., 5 km), and semantic similarity (cosine similarity > θ, e.g., 0.75).

[0055] For example, if multiple work orders have an element coverage rate >70%, meet the spatiotemporal proximity criteria, and are semantically similar, they are aggregated into a candidate set of "suspected new events" (e.g., multiple work orders are merged into "Chaoyang District power grid failure event"), and an evidence package (including element summary and original text link) is output for each event. This process ensures that "new words" are mapped to "new events," improves interpretability, and traces high-scoring new words back to the original work orders to form a candidate set of "suspected new events."

[0056] This embodiment overcomes the following shortcomings: "New word discovery" mainly relies on sudden increases in word frequency and a single statistical measure (such as Z-score) for judgment. In the initial stage when the number of work orders is small or when the first 1-3 words appear, the score fluctuates greatly and is easily amplified by random noise, leading to unstable early alarms. It is difficult to distinguish between one-time peaks and continuous peaks. The pure sliding window lacks the ability to distinguish between "occasional spikes" and "continuous increases," making it impossible for the operations side to determine whether follow-up is necessary. It is highly sensitive to word segmentation and parameters. Small changes in segmentation errors, window size, and threshold strategies can significantly affect the results, and its robustness is insufficient under unconventional distributions such as holidays and extreme weather.

[0057] The triggering time is delayed; most statistical methods require a significant threshold to trigger a signal, making it difficult to output stable judgments at the "first call" stage and failing to meet the "first-time" response requirements. Inconsistent cross-regional baselines and differences in call volume across regions lead to "strong regions masking weak regions," easily obscuring small-scale but high-risk new events, indicating insufficient normalization and adaptive baseline construction. A single ranking method is insufficient to guide handling priorities; ranking solely based on "new word scores" fails to balance credibility and business urgency, easily prioritizing "hot but not urgent" words. Interpretability and auditability are poor; evidence granularity remains at the statistical score of "words," lacking structured elements and visual evidence for business review, resulting in high review costs. Threshold parameter tuning relies on human experience; different holidays, weather, or policy windows cause frequent threshold failures, requiring repeated manual adjustments. Dictionaries, rules, and thresholds require continuous manual optimization, and performance regressions are prone to occur when migrating across quarters or themes. The traditional approach cannot support sophisticated screening at the operational level. It only provides a "full list," lacking the ability to filter by time window, theme, confidence interval, or other dimensions. The metrics only reflect the "novelty of the words," lacking quantitative evidence of "whether they are worth watching." Traditional scores make it difficult for staff to focus their limited time on words more likely to generate actionable value.

[0058] It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases the steps shown or described may be executed in a different order than that shown here.

[0059] According to an embodiment of the present invention, a device for identifying new words and first-time events is also provided, comprising: a preprocessing unit, configured to acquire hotline work order text data at a preset period and preprocess the hotline work order text data to obtain a structured text stream; a term candidate set determination unit, configured to perform word segmentation and sliding window processing on the text stream to obtain an initial phrase candidate set; perform statistical admission screening on the initial phrase candidate set, and perform alias merging on the candidate phrases that pass the statistical admission using a consistency rule of a specified dimension to obtain a term candidate set; a calibration unit, configured to update the current spatiotemporal reference baseline based on the term candidate set, and recalculate the entire historical data at a specified fixed period to recalibrate the spatiotemporal reference baseline; and a new word and first-time event determination unit, configured to determine whether there are new words in the term candidate set and whether there are first-time events based on the recalibrated spatiotemporal reference baseline.

[0060] According to embodiments of the present invention, the present invention also provides an electronic device, the electronic device comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to implement the methods described in any of the above embodiments.

[0061] According to embodiments of the present invention, the present invention also provides a readable storage medium storing computer instructions that enable a computer to perform the methods described in any of the above embodiments when executed.

[0062] According to embodiments of the present invention, the present invention also provides a computer program product that, when executed by a processor, can implement the methods described in any of the above embodiments.

[0063] Figure 2 A schematic block diagram of an example electronic device 300 that can be used to implement embodiments of the present invention is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device may also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices, and other similar computing devices.

[0064] like Figure 2As shown, the electronic device 300 includes a computing unit 301, which can perform various appropriate actions and processes based on a computer program stored in a read-only memory (ROM) 302 or a computer program loaded from a storage unit 308 into a random access memory (RAM) 303. The RAM 303 may also store various programs and data required for the operation of the electronic device 300. The computing unit 301, ROM 302, and RAM 303 are interconnected via a bus 304. An input / output (I / O) interface 305 is also connected to the bus 304.

[0065] Multiple components in electronic device 300 are connected to I / O interface 305, including: input unit 306, such as keyboard, mouse, etc.; output unit 307, such as various types of displays, speakers, etc.; storage unit 308, such as disk, optical disk, etc.; and communication unit 309, such as network card, modem, wireless transceiver, etc. Communication unit 309 allows electronic device 300 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0066] The computing unit 301 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the computing unit 301 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. The computing unit 301 performs the various methods and processes described above, such as the object matching method. For example, in some embodiments, the object matching method may be implemented as a computer software program tangibly contained in a machine-readable medium, such as storage unit 308. In some embodiments, part or all of the computer program may be loaded and / or installed on the electronic device 300 via ROM 302 and / or communication unit 309. When the computer program is loaded into RAM 303 and executed by the computing unit 301, one or more steps of the methods described above may be performed.

[0067] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-a-chip (SoCs), payload-programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0068] The program code used to implement the methods of the present invention can be written in any combination of one or more programming languages. This program code can be provided to a processor or controller of a general-purpose computer, special-purpose computer, or other programmable data processing device, such that when executed by the processor or controller, the program code causes the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The program code can be executed entirely on the machine, partially on the machine, as a standalone software package partially on the machine and partially on a remote machine, or entirely on a remote machine or server.

[0069] In the context of this invention, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. Machine-readable media can include, but are not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

Claims

1. A method for identifying new words and their initial occurrence events, characterized in that, include: Hotline work order text data is acquired according to a preset cycle, and the hotline work order text data is preprocessed to obtain a structured text stream. The text stream is segmented and processed using a sliding window method to obtain an initial set of candidate phrases; Statistical admission screening is performed on the initial phrase candidate set, and the candidate phrases that pass the statistical admission are merged by aliasing according to the consistency rules of the specified dimension to obtain the term candidate set; The current spatiotemporal reference baseline is updated based on the term candidate set, and the entire historical data is recalculated at a specified fixed period to recalibrate the spatiotemporal reference baseline. Based on the recalibrated spatiotemporal reference baseline, determine whether there are new words in the candidate word set and whether there is a first occurrence event; The process of determining whether new words exist in the candidate word set based on a recalibrated spatiotemporal reference baseline includes: outputting the expected probability and fluctuation range of words with a unified caliber in time and space based on the recalibrated spatiotemporal reference baseline; calculating a preset index value based on the expected probability and fluctuation range; calculating a new word index based on the preset index value; calculating a small sample confidence score using the Wilson interval estimation method with Bayesian prior for the candidate word set, and determining the trend persistence of words using an exponentially weighted moving average algorithm with hysteresis; and determining words in the candidate word set that belong to new words based on the new word index, small sample confidence score, and trend persistence. When determining whether a first event exists, the method includes: performing time series decomposition on the recalibrated spatiotemporal reference baseline to extract and remove holiday rhythms and diurnal rhythm components to obtain a residual baseline; among the new words identified, selecting words that have never appeared in the historical residual baseline and appear for the first time in the current window, and adding such words to the first event candidate pool; when the first event candidate words in the first event candidate pool meet the set conditions, a first event alarm is triggered.

2. The method for identifying new words and first-time events according to claim 1, characterized in that, The alias merging of candidate phrases that have passed the statistical admission criteria is performed using consistency rules of a specified dimension, including: calculating the homophonic and semantic similarity, as well as the difference between left and right entropy, for any two candidate phrases; If the similarity of homophones and near-homophones is greater than the first threshold, the semantic similarity is greater than the second threshold, and the left and right entropy values ​​are less than the third threshold, then the merging is performed.

3. The method for identifying new words and first-time events according to claim 1, characterized in that, Calculate a preset index value based on the expected probability of the term and the fluctuation range; The calculation of the new word index based on the preset index value includes: based on Calculate the new word index N Where Coh is the cohesion index, used to measure the cohesion within subsequent words; Div is the diversity index, used to suppress pseudo-words that are randomly spliced ​​together; Lift is the relative improvement index, determined by the ratio of the expected probability of words in the current window to the proportion of expected occurrences in the past; and Fresh is the time freshness index, calculated by normalizing the increase of the last k windows and the time decay term.

4. The method for identifying new words and first-time events according to claim 3, characterized in that, When the initial candidate terms in the initial candidate pool meet the set conditions, an initial alert is triggered, including: calculating the significance deviation value of the term, wherein the significance deviation value is used to represent the degree of standardized deviation of the term's occurrence probability from the historical baseline. , and For stratified historical mean and standard deviation, It is a smoothing factor; Calculate the short-term increase value of the term, where the short-term increase value represents the increase in the probability of the term appearing in the current window relative to the average of the last k windows: , It is the smoothed mean of a number of windows; If the significant deviation value meets the first threshold and the short-term increase meets the second threshold, an initial warning will be triggered.

5. The method for identifying new words and first-time events according to claim 4, characterized in that, When the initial candidate terms in the initial candidate pool meet the set conditions, an initial alert is triggered, including: The risk level of each term is determined based on a pre-defined risk term list, and a continuous score of 0-1 is output as the risk weight. If the risk weight hits a risk gate with a risk weight greater than the third threshold, the first alarm will be triggered.

6. The method for identifying new words and first-time events according to claim 5, characterized in that, The method also includes: The comprehensive score of the first-time event is calculated based on P = N×Conf×Persist×R×E, where N is the new word index, Conf is the small sample confidence level, Persist is the trend persistence, R is the risk weight, and E is the evidence level. The new words that triggered the first-time event are ranked based on the comprehensive score.

7. The method for identifying new words and first-time events according to claim 6, characterized in that, The system backtracks the original work order data for the sorted new words; extracts information of a specified dimension from the original work order data; determines the degree of evidence based on the specified dimension information and outputs an evidence package.

8. A device for identifying new words and their initial occurrence events, characterized in that, Includes: a preprocessing unit, used to acquire hotline work order text data at a preset period, and preprocess the hotline work order text data to obtain a structured text stream; The term candidate set determination unit is used to perform word segmentation and sliding window processing on the text stream to obtain an initial term candidate set; Statistical admission screening is performed on the initial phrase candidate set, and the candidate phrases that pass the statistical admission are merged by aliasing according to the consistency rules of the specified dimension to obtain the term candidate set; The calibration unit is used to update the current spatiotemporal reference baseline based on the term candidate set, and to recalculate the entire historical data at a specified fixed period in order to recalibrate the spatiotemporal reference baseline. The new word and first-time event determination unit is used to determine whether there are new words in the word candidate set and whether there are first-time events based on the recalibrated spatiotemporal reference baseline. The process of determining whether new words exist in the candidate word set based on a recalibrated spatiotemporal reference baseline includes: outputting the expected probability and fluctuation range of words with a unified caliber in time and space based on the recalibrated spatiotemporal reference baseline; calculating a preset index value based on the expected probability and fluctuation range; calculating a new word index based on the preset index value; calculating a small sample confidence score using the Wilson interval estimation method with Bayesian prior for the candidate word set, and determining the trend persistence of words using an exponentially weighted moving average algorithm with hysteresis; and determining words in the candidate word set that belong to new words based on the new word index, small sample confidence score, and trend persistence. When determining whether a first event exists, time series decomposition is performed on the recalibrated spatiotemporal reference baseline to extract and remove holiday rhythms and diurnal rhythm components to obtain the residual baseline; among the new words identified, words that have never appeared in the historical residual baseline and appear for the first time in the current window are selected and added to the first event candidate pool; when the first event candidate words in the first event candidate pool meet the set conditions, a first event alarm is triggered.

Images