Information tracing method and device, electronic equipment and storage medium

Abstract

This invention relates to the field of information tracing technology, and discloses information tracing methods, devices, electronic devices, and storage media. The invention first disassembles internet information to obtain combined features formed by the combination of event subject characteristics and event description characteristics, along with corresponding time parameters, constructing a dynamically searchable disassembled feature library. After acquiring the information to be traced, based on its event subject, event description, and event occurrence time, the disassembled feature library is retrieved, and a query feature cluster is constructed based on the retrieved combined features. A thorough tracing search is then performed based on the query feature cluster. The earliest publication time of the traceable content is compared with the currently recorded tracing time. If the earliest publication time is earlier than the tracing time, the tracing time is updated to the earliest publication time, and the feature information of the traceable content is added to the information to be traced, for the next iteration. In this way, each iteration supplements earlier features and new keywords, thus effectively tracing back to an earlier, true initial source.

Description

Technical Field

[0001] This invention relates to the field of information traceability technology, specifically to information traceability methods, devices, electronic equipment, and storage media. Background Technology

[0002] First-release information tracing is an activity that tracks the original source, first release time, and initial content of information, and plays an important role in public opinion control and information retrieval.

[0003] The current traditional method for tracing the first source of information generally involves manually extracting the main descriptive sentences of the event or using methods such as Term Frequency-Inverse Document Frequency (TF-IDF) to extract the keywords of the text, and then searching in a full-text search database to obtain the earliest published information as the first source of information.

[0004] However, in the era of self-media, especially in recent years, the source of internet dissemination has undergone new changes. It first appears in more colloquial forms such as short videos, documents, audio recordings, and comments on social platforms, and then is widely disseminated through briefings and processing by news media and influential figures. Traditional methods can only trace the source based on keywords that have already appeared in the input text, making it difficult to obtain the original self-media form before the change. Summary of the Invention

[0005] This invention provides an information tracing method, apparatus, electronic device, and storage medium to address the problem that the diversification of the initial information formats makes it difficult to obtain the original initial information using traditional methods.

[0006] In a first aspect, the present invention provides an information tracing method, the method comprising:

[0007] Construct a feature decomposition library, which includes multiple combined features and time parameters for each combined feature. The combined features include the event subject features and event description features of the corresponding Internet information.

[0008] Once the information to be traced is determined, the event subject, event description, and event occurrence time of the information to be traced are retrieved and decomposed from the feature database to obtain the query feature clusters.

[0009] The source search is performed based on the query feature clusters to obtain the source content, determine the earliest time of the source content, and obtain the source time of the current record.

[0010] If the earliest sent time is detected to be earlier than the source tracing time, the source tracing time is updated, and the source information to be traced is updated based on the source tracing content. The process returns to the step of determining the source information to be traced, until the earliest sent time is no earlier than the source tracing time, and the source tracing result is obtained based on the source tracing content.

[0011] This invention first disassembles internet information to obtain combined features formed by the combination of event subject characteristics and event description characteristics, along with corresponding time parameters, thereby constructing a dynamically updated and queryable disassembled feature library. After obtaining the information to be traced, the disassembled feature library is retrieved based on its event subject, event description, and event occurrence time, and a query feature cluster is constructed based on the retrieved combined features. A thorough source tracing search is then performed based on the query feature cluster. The earliest publication time of the searched source content is compared with the currently recorded source tracing time. If the earliest publication time is earlier than the source tracing time, it indicates that there may be an earlier initial source. The source tracing time is then updated to the earliest publication time, and the feature information of the source content is added to the information to be traced, for the next iteration. In this way, each iteration supplements earlier features and new keywords, thus effectively tracing back to an earlier, true initial source.

[0012] In one optional implementation, the traceability content includes at least some of the following: text content, image content, audio content, and video content; updating the traceability information based on the traceability content includes:

[0013] Extract multimodal feature information from the source content;

[0014] Add multimodal feature information to the information to be traced to obtain the updated information to be traced.

[0015] This invention addresses the issue where, before the iteration ends (i.e., the earliest recorded time retrieved in the current iteration is earlier than the original source), features from the text, images, audio, or video in the source content are extracted and added to the source information to be traced, allowing for the next iteration. Each iteration adds earlier features and new keywords, effectively tracing back to the earliest true original source.

[0016] In one optional implementation, the time parameters include the earliest occurrence date, the number of occurrences on the first occurrence date, the outbreak time, and the independence weight; based on the event subject, event description, and event occurrence time of the information to be traced, the feature library is retrieved and decomposed to obtain the query feature cluster, including:

[0017] Based on the event subject and event description of the information to be traced, a combination feature to be traced is constructed, and a feature library is retrieved and decomposed based on the combination feature to be traced to obtain multiple candidate combination features;

[0018] Determine the source tracing period and divide it into multiple time windows;

[0019] Based on the number of times the first day of occurrence, the time of suddenness, and the independence weight of the candidate combination features within each time window, a comprehensive score for each time window is obtained, and the target time window with the highest comprehensive score is determined.

[0020] Based on the event occurrence time, a secondary screening of candidate combination features within the target time window is performed to obtain the query feature cluster.

[0021] This invention generalizes the event subject and event description to construct a set of features to be traced, and then fully searches and decomposes the feature library to obtain multiple matching candidate features. The traceability period is divided into multiple time windows. Based on the frequency of occurrence on the first day of the candidate features, the time of sudden occurrence, and the independence weight, each time window is scored, and the target time window with the highest comprehensive score is determined to initially filter irrelevant features. A second screening is then performed on the candidate features within the target time window to further filter irrelevant information, constructing a query feature cluster that better matches the information to be traced.

[0022] In one optional implementation, candidate combination features within the target time window are further filtered based on the event occurrence time to obtain a query feature cluster, including:

[0023] Based on the earliest occurrence date of each candidate combination feature within the target time window, the earliest occurrence date with the most occurrences is determined as the maximum co-occurrence date;

[0024] If the earliest occurrence date of any candidate combination feature is detected to be earlier than the maximum co-occurrence date by at least a certain number of days, the candidate combination feature will be removed from the target time window.

[0025] If the earliest occurrence date of any candidate combination feature is detected to be earlier than the event occurrence time, and the number of occurrences of the first day of the candidate combination feature is higher than the occurrence threshold, the candidate combination feature will be removed from the target time window.

[0026] Based on the remaining candidate combination features within the target time window, the query feature cluster is obtained.

[0027] This invention removes candidate combination features that frequently appear before the event occurs, and removes candidate combination features whose earliest occurrence date deviates too much from the maximum co-occurrence date. In this way, a query feature cluster is constructed based on the remaining candidate combination features within the target time window, ensuring that all query features are aligned in time and strongly correlated with the information to be traced, thus avoiding the retrieval of invalid traceability content.

[0028] In one optional implementation, based on the event subject and event description of the information to be traced, a combined feature to be traced is constructed, including:

[0029] The event subject and event description are generalized to construct a generalized feature library; the generalized feature library includes multiple generalized subjects and multiple generalized descriptions;

[0030] By combining any generalization subject and generalization description in the generalization feature library, the source-tracing combined features are obtained.

[0031] This invention generalizes the event subject and event description of the information to be traced into colloquial language to obtain corresponding generalized subjects and generalized descriptions. By arbitrarily combining the generalized subjects and generalized descriptions, the combined features to be traced are obtained, which cover more colloquial expressions and avoid missing traceability information.

[0032] In one alternative implementation, the method further includes:

[0033] Cross-validate whether each query feature in a query feature cluster belongs to the same event;

[0034] If outlier query features that do not belong to the same event are identified, retrieve the target Internet information corresponding to the time of the event, and verify the outlier query features based on the target Internet information;

[0035] If the outlier query feature fails the validation, the outlier query feature will be removed from the query feature cluster.

[0036] This invention cross-determines whether each query feature in a query feature cluster belongs to the same event. If an outlier query feature that does not belong to the same event is identified, the outlier query feature is verified using the target Internet information that actually occurred on the day the event occurred. If the verification fails, it is removed from the query feature cluster, thereby ensuring the validity and accuracy of the query features.

[0037] In one optional implementation, cross-validating whether each query feature in a query feature cluster belongs to the same event includes:

[0038] Retrieve the list of source addresses corresponding to each query feature;

[0039] If overlapping source addresses are detected in the source address lists of any number of query features, then the number of query features is determined to belong to the same event.

[0040] This invention constructs a source address list by retrieving multiple source addresses corresponding to each query feature. It then determines whether there are overlapping source addresses between these lists, i.e., whether a certain source address appears simultaneously in multiple source address lists corresponding to different query features. If so, the multiple query features are determined to belong to the same event. This allows for rapid same-event determination without needing to obtain source information corresponding to the query features and calculate semantic similarity, significantly improving the system's source tracing efficiency.

[0041] Secondly, the present invention provides an information tracing device, the device comprising:

[0042] The first processing module is used to construct a decomposed feature library. The decomposed feature library includes multiple combined features and time parameters for each combined feature. The combined features include the event subject features and event description features of the corresponding Internet information.

[0043] The second processing module is used to determine the information to be traced, and based on the event subject, event description and event occurrence time of the information to be traced, it retrieves and decomposes the feature library to obtain the query feature cluster;

[0044] The third processing module is used to perform source tracing search based on query feature clusters, obtain source content, determine the earliest time of the source content, and obtain the source tracing time of the current record.

[0045] The fourth processing module is used to update the tracing time if the earliest sent time is detected to be earlier than the tracing time, and to update the tracing information to be traced based on the tracing content, and return to the step of determining the tracing information to be traced, until the earliest sent time is no earlier than the tracing time, and the tracing result is obtained based on the tracing content.

[0046] Thirdly, the present invention provides an electronic device, comprising: a memory and a processor, wherein the memory and the processor are communicatively connected to each other, the memory stores computer instructions, and the processor executes the computer instructions to perform the information tracing method described in the first aspect or any corresponding embodiment thereof.

[0047] Fourthly, the present invention provides a computer-readable storage medium storing computer instructions for causing a computer to execute the information tracing method described in the first aspect or any corresponding embodiment thereof. Attached Figure Description

[0048] 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.

[0049] Figure 1 This is a schematic diagram of an application scenario according to an embodiment of the present invention;

[0050] Figure 2 This is a schematic diagram of the first type of information tracing method according to an embodiment of the present invention;

[0051] Figure 3 This is a schematic diagram of the second type of information tracing method according to an embodiment of the present invention;

[0052] Figure 4 This is a schematic diagram of the third process of the information tracing method according to an embodiment of the present invention;

[0053] Figure 5 This is a structural block diagram of an information tracing device according to an embodiment of the present invention;

[0054] Figure 6 This is a schematic diagram of the hardware structure of an electronic device according to an embodiment of the present invention. Detailed Implementation

[0055] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, 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, 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.

[0056] It is understood that before using the technical solutions disclosed in the various embodiments of the present invention, users should be informed of the types, scope of use, and usage scenarios of the personal information involved in the present invention and their authorization should be obtained in accordance with relevant laws and regulations through appropriate means.

[0057] The terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0058] As an optional application scenario of this invention, such as Figure 1 As shown, the information traceability system may include at least one terminal device and at least one server. Figure 1The system is illustrated in the example, which includes a computer 101, a mobile terminal 102, and a server 103, and the terminal devices such as the computer 101 and the mobile terminal 102 are connected to the server 103 through a network 110.

[0059] Specifically, the terminal device can be a smartphone, tablet, laptop, PDA, desktop computer, game console, smart TV, smart wearable device, in-vehicle terminal, VR (Virtual Reality) device, AR (Augmented Reality) device, etc. Server 103 can be a standalone physical server, a server cluster, a distributed system, or a cloud server providing cloud services. Network 110 can be a wired or wireless network, examples of which include, but are not limited to, the Internet, corporate intranet, local area network, wide area network, mobile communication network, and combinations thereof.

[0060] Traditional methods for tracing the origin of information on the internet typically involve manually extracting key descriptive sentences of the event or using techniques like TF-IDF to extract keywords from the text, then searching a full-text database to find the earliest published information as the initial source. Clustering methods are also used, first performing clustering using methods like Simhash, and then selecting the earliest information from each cluster for tracing. However, these methods can only trace back based on features already present in the input text. News articles and self-media often have complex language patterns and contain a lot of information, often only finding the earliest information currently being disseminated by news or self-media, failing to achieve truly effective source tracing.

[0061] Another approach is to use a machine-trained feature combination fingerprint clustering method for tracing back to the source by using real-time features of third-order word co-occurrence through machine learning. However, this method has limited generalization ability and cannot iteratively find new generalized features for early information tracing.

[0062] This invention provides an information tracing method that continuously collects and characterizes internet information to construct a feature decomposition database. Based on the information to be traced, combined features from the feature decomposition database are selected to construct a query feature cluster. A full-text search is performed using the query features in the query feature cluster to obtain the traceable content. If the earliest publication time of the traceable content does not meet the iteration termination condition, the traceable content is used to supplement the information to be traced. Thus, each iteration supplements earlier features and new keywords, effectively tracing back to an earlier, true initial publication point.

[0063] According to an embodiment of the present invention, an embodiment of an information tracing method is provided. 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. Furthermore, 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.

[0064] This embodiment provides an information tracing method, which can be used in the aforementioned terminal devices, such as mobile phones and tablet computers. Figure 2 This is a flowchart of an information tracing method according to an embodiment of the present invention, such as... Figure 2 As shown, the process includes the following steps:

[0065] Step S201: Construct a decomposed feature library. The decomposed feature library includes multiple combined features and time parameters for each combined feature. The combined features include the event subject features and event description features of the corresponding Internet information.

[0066] Specifically, publicly available internet information is continuously collected, including but not limited to user comments, videos, audio, and images posted on news media, social media, and short video platforms. This internet information is clustered, and the information within each cluster is broken down to obtain the event subject characteristics and event description characteristics for each piece of internet information. The event subject characteristics and corresponding event description characteristics are combined to form combined features, and time parameters such as the earliest occurrence date, the number of times the first occurrence occurred, the time of suddenness, and the independence weight are collected for these combined features.

[0067] In this embodiment, the earliest occurrence date is obtained by collecting the sending time of the corresponding Internet information. The number of occurrences on the first occurrence date refers to the number of times the corresponding Internet information appears on the earliest occurrence date. The burst time refers to the point in time when the frequency of occurrence of similar Internet information increases sharply (e.g., the point in time when the number of Internet information in a cluster increases sharply).

[0068] In this embodiment, the independence weight is used to characterize the uniqueness of the combined feature. The independence weight can be obtained by calculating the correlation between the combined feature and other combined features. The stronger the correlation, the higher the degree of overlap between the combined feature and other combined features, and the lower the independence weight. Conversely, the weaker the correlation, the better the combined feature can distinguish specific events within the characteristic time period, and the higher the independence weight.

[0069] In some embodiments, all internet information from a designated media platform is categorized and extracted periodically (e.g., every 14 days or longer, adjustable according to actual needs). Event subject features (e.g., region, organization, person, product name, etc.) and event description features (event words, negative words, proper nouns, colloquial generalized descriptions, etc.) are extracted from each piece of internet information. Combined features are then constructed, and all combined features are hashed. The hash representation of the combined features, along with parameters such as the earliest occurrence date, the number of occurrences on the first day, the time of sudden appearance, and the independence weight, are recorded to form a dynamically updated decomposed feature library.

[0070] In some embodiments, Natural Language Processing (NLP) can be used to extract the event subject features and event description features of each piece of Internet information. For details, please refer to the description of the relevant technologies, which will not be repeated here.

[0071] Step S202: Determine the information to be traced. Based on the event subject, event description, and event occurrence time of the information to be traced, retrieve and decompose the feature database to obtain the query feature cluster.

[0072] Specifically, the system obtains the traceable information input by the user, such as the title, text, image, video, or audio of the information to be traced. First, it performs feature decomposition, breaking down the traceable information into: the event subject (region, organization, person, product name, etc.) and the event description (event words, negative words, proper nouns, colloquial generalized descriptions, etc.).

[0073] In some embodiments, if the information to be traced contains the event occurrence time, the event occurrence time of the corresponding event can be directly extracted; if the information to be traced does not contain the event occurrence time, the date and time of the event can be estimated by analyzing the large model and performing a simple search on the information to be traced, thereby obtaining the event occurrence time.

[0074] In this embodiment, the event subject, event description, and event occurrence time of the information to be traced are used as key retrieval information. The combined features in the feature decomposition library are filtered to obtain combined features that match the information to be traced. Query feature clusters are then constructed based on these combined features.

[0075] Step S203: Perform source tracing search based on query feature clusters to obtain source content, determine the earliest time of the source content, and obtain the source tracing time of the current record.

[0076] Specifically, the query feature cluster includes multiple query features. Based on each query feature, an Elasticsearch (ES) search is performed to aggregate the traceability information to obtain traceability content. The traceability content includes at least some of the following items: text content, image content, audio content, and video content.

[0077] In some embodiments, a large model is used to extract summaries of the searched source information by time, region, subject, and event description. The large model is then used to compare the summary content with the source information to be traced to confirm whether they are different descriptions of the same event. If they describe the same event, the source information is then summarized.

[0078] In this embodiment, the earliest issuance time of the traceable content is extracted and compared with the recorded traceability time. If the earliest issuance time is earlier than the traceability time, step S204 is executed; if the earliest issuance time is not earlier than the traceability time, the iteration ends. The traceability time refers to the initial time of the traceable information currently being traced. In the first iteration, the currently recorded traceability time can be the event occurrence time.

[0079] Step S204: If the earliest sent time is detected to be earlier than the source tracing time, update the source tracing time and update the source information to be traced based on the source tracing content. Return to the step of determining the source information to be traced until the earliest sent time is no earlier than the source tracing time, and obtain the source tracing result based on the source tracing content.

[0080] Specifically, if the earliest sent time is earlier than the tracing time, it means that the traced content is not the initial information and iteration needs to continue. In this case, the tracing time is updated to the earliest sent time of the traced content in the current iteration, and the feature information of the traced content is added to the information to be traced. Then, return to step S202 for the next round of iteration. During the iteration, if the earliest sent time is detected to be no earlier than the tracing time, it means that the traced content is already the earliest. In this case, the traced content can be identified as the tracing result, thereby determining the initial source of the information to be traced.

[0081] In the era of self-media, information is often initially presented in the form of short, fragmented, colloquial videos, comments, and audio recordings, before being transformed into standardized news language for dissemination. Traditional source tracing techniques (such as keyword retrieval, TF-IDF, and SimHash clustering) rely on the characteristics of the query text itself, making it difficult to effectively trace back to earlier, more diverse original information sources, resulting in inaccurate and incomplete tracing results.

[0082] The information tracing method provided in this embodiment first deconstructs internet information to obtain combined features formed by the combination of event subject features and event description features, along with corresponding time parameters, thereby constructing a dynamically updated and queryable deconstructed feature library. After obtaining the information to be traced, the deconstructed feature library is retrieved based on its event subject, event description, and event occurrence time, and a query feature cluster is constructed based on the retrieved combined features. A thorough tracing search is then performed based on the query feature cluster. The earliest publication time of the searched tracing content is compared with the currently recorded tracing time. If the earliest publication time is earlier than the tracing time, it indicates that there may be an earlier initial source. The tracing time is then updated to the earliest publication time, and the feature information of the tracing content is added to the information to be traced, for the next iteration. In this way, each iteration supplements earlier features and new keywords, thus effectively tracing back to an earlier, true initial source.

[0083] This embodiment provides an information tracing method, which can be used in the aforementioned terminal devices, such as mobile phones and tablet computers. Figure 3 This is a flowchart of an information tracing method according to an embodiment of the present invention, such as... Figure 3 As shown, the process includes the following steps:

[0084] Step S301: Construct a feature decomposition library. The feature decomposition library includes multiple combined features and time parameters for each combined feature. The combined features include the event subject features and event description features of the corresponding internet information. For details, please refer to... Figure 2 Step S201 of the illustrated embodiment will not be described again here.

[0085] Step S302: Determine the information to be traced. Based on the event subject, event description, and event occurrence time of the information to be traced, retrieve and decompose the feature database to obtain the query feature cluster.

[0086] Specifically, step S302 includes:

[0087] Step S3021: Based on the event subject and event description of the information to be traced, construct the combined features to be traced, and retrieve the decomposed feature library based on the combined features to be traced to obtain multiple candidate combined features.

[0088] In some optional implementations, step S3021 above includes:

[0089] Step a1: Generalize the event subject and event description to construct a generalized feature library; wherein, the generalized feature library includes multiple generalized subjects and multiple generalized descriptions.

[0090] Specifically, since online self-media and news expressions tend to be more formal, these formal expressions may not necessarily be the same as those used in the original article. Therefore, the subject and description of the event are expanded. The subject and description of the event are generalized using word2vec to achieve a colloquial style, resulting in multiple generalized subjects and descriptions, which are then used to construct a generalized feature library.

[0091] Taking the information to be traced as "May 14, video shows meteor shower, another picture shows famous landmark building in Jinjiang" as an example, the subject of the event is "Jinjiang", and the event description is "Leonid meteor shower". Then the generalized feature library can be shown in Table 1 below:

[0092] Table 1 Generalization Feature Library

[0093] Generalized subject Generalized subject relevance Generalized description Generalized description of relevance JinJ District 0.9243 Meteor Shower 0.7663 WuH District 0.9207 meteor 0.7593 Gold N District 0.9202 Leo 0.7527 Chengdu D City 0.8811 meteorite fragments 0.7467 New D District 0.8555 rain 0.7220

[0094] Step a2: Combine any generalization subject and generalization description in the generalization feature library to obtain the source-tracing combined feature.

[0095] Specifically, any generalized subject and any generalized description in the generalization feature library are paired to obtain multiple source-tracing combined features. Then, based on the source-tracing combined features, the feature library is decomposed to obtain multiple matching candidate combined features.

[0096] This embodiment generalizes the event subject and event description of the information to be traced into colloquial language to obtain the corresponding generalized subject and generalized description. By arbitrarily combining the generalized subject and generalized description, the combined features to be traced are obtained, so that the combined features to be traced cover more colloquial expressions and avoid missing the traceability information.

[0097] Step S3022: Determine the source tracing time period and divide the source tracing time period into multiple time windows.

[0098] Specifically, the earliest occurrence date corresponding to each candidate combination feature is determined, and the time period covered by the earliest occurrence date is defined as the source tracing time period. A preset time window (e.g., half a year, 3 months, etc.) is slid across the source tracing time period to divide the source tracing time period into multiple time windows.

[0099] Step S3023: Based on the number of times the first day of occurrence, the suddenness and independence weight of the candidate combination features within each time window, obtain the comprehensive score for each time window, and determine the target time window with the largest comprehensive score.

[0100] Specifically, for each candidate combination feature within the time window, scores are assigned and weighted sums are performed based on the number of times the initial date occurs, the timing of the outbreak, and the independence weight to calculate the score of the corresponding candidate combination feature. It should be noted that the specific scoring can be set according to the actual scenario requirements; the more times the initial date occurs, the closer the timing of the outbreak is to the event occurrence time, and the higher the independence weight, the higher the score of the candidate combination feature.

[0101] Furthermore, for each time window, the scores of all candidate combination features within that time window are summed and normalized to obtain a comprehensive score for that time window. The time window with the highest comprehensive score is then determined as the target time window, where the candidate combination features are considered to be closely related to the source information to be traced.

[0102] In step S3024, the candidate combination features within the target time window are further filtered based on the event occurrence time to obtain the query feature cluster.

[0103] In some optional implementations, step S3024 above includes:

[0104] Step b1: Based on the earliest occurrence date of each candidate combination feature within the target time window, determine the earliest occurrence date with the most occurrences as the maximum co-occurrence date.

[0105] For example, the candidate combined features available within the target time window can be:

[0106] The first appearance of the D-type meteor shower was on May 14, 2025, and it appeared 17 times on its first appearance day.

[0107] The earliest appearance date of the "D AND meteor" trait is March 11, 2025. The number of times it appeared on the first day of its appearance is 344.

[0108] The first appearance date of the combination "D AND Leo" is May 1, 2025, and the number of times it appears on the first day of its appearance is 73.

[0109] JinJAND Rain, first appeared on December 19, 2024, appeared 3 times on the first day;

[0110] JinJAND Meteor, first appeared on May 4, 2025, appeared 7 times on its first day;

[0111] JinJAND Leo, first appeared on May 14, 2025, first appearance date: 19;

[0112] The JinJAND meteor shower first appeared on May 14, 2025, and appeared 22 times on its first day.

[0113] Following the example above, the earliest date of occurrence with the highest frequency is 20250514, so 20250514 is taken as the date of maximum co-occurrence.

[0114] Step b2: If the earliest occurrence date of any candidate combination feature is detected to be earlier than the maximum co-occurrence date by at least a number of days, the candidate combination feature is removed from the target time window.

[0115] Specifically, each candidate combination feature within the target time window is examined. If the earliest occurrence date of a candidate combination feature is more than a threshold number of days earlier than the maximum co-occurrence date, it is determined to be an invalid feature and removed from the target time window. The threshold number of days can be 2 days and is adjustable.

[0116] Step b3: If the earliest occurrence date of any candidate combination feature is detected to be earlier than the event occurrence time, and the number of occurrences of the first day of the candidate combination feature is higher than the occurrence threshold, the candidate combination feature is removed from the target time window.

[0117] Specifically, if the earliest occurrence date of a candidate feature combination is earlier than the event occurrence time, and the number of occurrences on the first day exceeds the occurrence threshold, it is removed from the target time window. The occurrence threshold can be 9, and can be adjusted according to actual needs.

[0118] Step b4: Based on the remaining candidate combination features within the target time window, obtain the query feature cluster.

[0119] Following the example above, the remaining candidate combination features within the target time window are:

[0120] The first appearance of the D-type meteor shower was on May 14, 2025, and it appeared 17 times on its first appearance day.

[0121] JinJAND Rain, first appeared on December 19, 2024, appeared 3 times on the first day;

[0122] JinJAND Meteor, first appeared on May 4, 2025, appeared 7 times on its first day;

[0123] JinJAND Leo, first appeared on May 14, 2025, first appearance date: 19;

[0124] The JinJAND meteor shower first appeared on May 14, 2025, and appeared 22 times on its first day.

[0125] Specifically, by excluding and filtering candidate combination features within the target window, a query feature cluster is constructed based on the remaining candidate combination features within the target time window, namely, candidate combination features that are sudden and have high independence weights near the time of the event.

[0126] This embodiment removes candidate combination features that frequently appeared before the event occurred, and removes candidate combination features whose earliest occurrence date deviates too much from the maximum co-occurrence date. In this way, a query feature cluster is constructed based on the remaining candidate combination features within the target time window, ensuring that all query features are aligned in time and strongly correlated with the information to be traced, thus avoiding the retrieval of invalid traceability content.

[0127] In the above embodiments, by generalizing the event subject and event description, a combination of features to be traced is constructed. Then, the feature library is fully searched and decomposed to obtain multiple matching candidate combination features. The traceability period is divided into multiple time windows. Based on the number of times the candidate combination features appear on their first day, the time of sudden occurrence, and the independence weight, each time window is scored. The target time window with the highest comprehensive score is determined, initially filtering out irrelevant combination features. A second screening is then performed on the candidate combination features within the target time window to further filter out irrelevant information, constructing a query feature cluster that better matches the information to be traced.

[0128] In some optional implementations, after obtaining the query feature clusters, steps c1 to c3 are performed:

[0129] Step c1: Cross-validate whether each query feature in the query feature cluster belongs to the same event.

[0130] Specifically, obtain the source address list corresponding to each query feature. If overlapping source addresses are detected in the source address lists of any number of query features, then it is determined that any number of query features belong to the same event.

[0131] In this embodiment, each query feature is used for retrieval to obtain a URL list corresponding to each query feature. This URL list contains multiple URL addresses. It is determined whether there is overlap between the URL lists corresponding to each query feature. If overlapping source addresses are detected between any multiple URL lists, that is, a certain URL appears in multiple URL lists at the same time, it is determined that the corresponding multiple query features belong to the same event.

[0132] This embodiment constructs a source address list by retrieving multiple source addresses corresponding to each query feature. It then determines whether there are overlapping source addresses between the source address lists, i.e., whether a certain source address appears simultaneously in multiple source address lists corresponding to multiple query features. If so, it determines that the corresponding multiple query features belong to the same event. This allows for rapid same-event determination without needing to obtain the source information corresponding to the query features and calculate semantic similarity, significantly improving the system's source tracing efficiency.

[0133] Step c2: If outlier query features that do not belong to the same event are identified, retrieve the target Internet information corresponding to the time of the event, and verify the outlier query features based on the target Internet information.

[0134] Specifically, if outlier query features that do not belong to the same event are identified through cross-validation, the target internet information for the day the event occurred is obtained. This target internet information may include the 10 most popular data items on the day the event occurred. The outlier query features are validated based on the target internet information to determine whether the corresponding fields of the outlier query features exist in the target internet information. If they exist, the validation passes, and the outlier query features are maintained in the query feature cluster.

[0135] Step c3: If the outlier query feature fails the validation, remove the outlier query feature from the query feature cluster.

[0136] In this embodiment, it is cross-checked whether each query feature in the query feature cluster belongs to the same event. If an outlier query feature that does not belong to the same event is identified, the outlier query feature is verified using the target Internet information that actually occurred on the day the event occurred. If the verification fails, it is removed from the query feature cluster, thereby ensuring the validity and accuracy of the query features.

[0137] In some embodiments, after obtaining the query feature cluster, the corresponding combination feature in the decomposed feature library for each query feature is determined, the earliest occurrence date of each combination feature is compared with the event occurrence time, the independence weight of the combination feature earlier than the event occurrence time is reduced, and the independence weight of the combination feature with the same event occurrence time is increased, thereby updating the parameters of the decomposed feature library.

[0138] Step S303: Perform a source tracing search based on the query feature clusters to obtain the source content, determine the earliest issuance time of the source content, and obtain the source tracing time of the current record. For details, please refer to [link / reference]. Figure 2 Step S303 of the illustrated embodiment will not be described again here.

[0139] Step S304: If the earliest sent time is detected to be earlier than the source tracing time, update the source tracing time and update the source information to be traced based on the source tracing content. Return to the step of determining the source information to be traced until the earliest sent time is no earlier than the source tracing time, and obtain the source tracing result based on the source tracing content.

[0140] Specifically, step S304 includes:

[0141] Step S3041: If the earliest sent time is detected to be earlier than the source tracing time, update the source tracing time.

[0142] Specifically, if the earliest issuance time of the current iteration step is detected to be earlier than the source time, the source time is updated to the earliest issuance time.

[0143] Step S3042: Extract multimodal feature information of the source content, add multimodal feature information to the source information to be traced, and obtain the updated source information to be traced.

[0144] In some embodiments, after obtaining the source content of the current iteration step, the earliest 5 pieces of information (including text, images, videos, and audio) in the source content are taken, and the corresponding multimodal feature information (including text keywords, image OCR information, video OCR information, and audio features) is identified. New subject and event description features are extracted from these features, and combined to obtain new combined features. These new combined features are then added to the source information to be traced, and the next round of iterative source tracing is performed.

[0145] Step S3043: Return to step S302 until the earliest issuance time is no earlier than the source tracing time, and obtain the source tracing result based on the source tracing content.

[0146] Specifically, return to step S302, perform source tracing iterations based on the new source information to be traced, until the source tracing time is no longer updated, and take the source tracing content as the final source tracing result, thereby determining the original source.

[0147] For example, the current source information is "On May 14th, the Leonid meteor shower appeared in Jinjiang," and the OCR information of the accompanying image is "XX Department Store." The extracted new feature "XX Department Store" is added to the source information to be traced, and the process returns to step S302 for another iteration. Then, new source information is obtained. If no new earliest information is added after two rounds of tracing, the tracing result is output and the iteration ends. After the iteration ends, the earliest published information is obtained, which is the initial post information. For example, the initial post information might be a user's post on a social media platform saying "The meteor shower can be seen at XX Department Store."

[0148] The information tracing method provided in this embodiment, if the tracing process does not reach the iteration termination condition (i.e., the earliest publication time retrieved in the current iteration is earlier than the tracing time), extracts features from the text, images, audio, or video in the tracing content, and supplements the tracing information with the extracted multimodal feature information, proceeding to the next round of tracing iteration. In this way, each iteration supplements earlier features and new keywords, thus effectively tracing back to an earlier, true initial source.

[0149] The information traceability scheme of the present invention will be described in detail below with reference to a specific application example.

[0150] like Figure 4 As shown, the first step is to perform overall combined training on Internet information data, classify and extract all information within a period of 14 days or longer, extract the main features and descriptive features of the events in the information, construct combined features, hash all combined features, and record the time parameters of all combined features to form a pre-trained decomposed feature library.

[0151] After the feature library is constructed, the information to be traced is obtained. The approximate time of the event is calculated using a large model. Then, the information to be traced is broken down into the event subject and event description. The event subject and description are generalized to colloquial language and searched in the feature library to obtain multiple candidate feature combinations. Multiple time windows are then created, and the time parameters of the candidate feature combinations are used to score each time window, calculating the target time window with the highest comprehensive score. The available candidate feature combinations within the target time window are then further filtered to construct a query feature cluster. A full-text search is performed using the query feature cluster for the first round of source tracing. Simultaneously, the large model extracts summaries of the traced content based on time, region, subject, and event. These summaries are then compared with the information to be traced to confirm whether they are different descriptions of the same event.

[0152] After obtaining the first round of source tracing results, the iterative source tracing process begins. It checks whether the earliest time of the source content was sent is earlier than the currently recorded source tracing time. If it is earlier, it means that earlier information has been discovered, and new features are extracted from the source content and added to the source tracing information for the next round of source tracing. If it is not earlier, the source content found in the current iteration is output as the source tracing result, and the iteration ends.

[0153] This invention utilizes more colloquial generalization features and iterative new feature extraction to trace the source of Internet information. By supplementing details through multiple iterations to complete the event description before tracing the source of the initial information, it can effectively trace back to the earlier true source.

[0154] This embodiment also provides an information tracing device for implementing the above embodiments and preferred embodiments; details already described will not be repeated. As used below, the term "module" can refer to a combination of software and / or hardware that performs a predetermined function. Although the device described in the following embodiments is preferably implemented in software, hardware implementation, or a combination of software and hardware, is also possible and contemplated.

[0155] This embodiment provides an information traceability device, such as... Figure 5 As shown, it includes:

[0156] The first processing module 501 is used to construct a decomposition feature library. The decomposition feature library includes multiple combined features and time parameters for each combined feature. The combined features include event subject features and event description features of the corresponding Internet information.

[0157] The second processing module 502 is used to determine the information to be traced, and based on the event subject, event description and event occurrence time of the information to be traced, to retrieve and decompose the feature library to obtain the query feature cluster;

[0158] The third processing module 503 is used to perform source tracing search based on query feature clusters, obtain source content, determine the earliest time of the source content, and obtain the source tracing time of the current record.

[0159] The fourth processing module 504 is used to update the tracing time if the earliest sent time is detected to be earlier than the tracing time, and update the tracing information to be traced based on the tracing content, and return to the step of determining the tracing information to be traced, until the earliest sent time is no earlier than the tracing time, and obtain the tracing result based on the tracing content.

[0160] In some optional implementations, the time parameters include the earliest occurrence date, the number of occurrences on the first occurrence date, the burst time, and the independence weight; the second processing module 502 is also used for:

[0161] Based on the event subject and event description of the information to be traced, a combination feature to be traced is constructed, and a feature library is retrieved and decomposed based on the combination feature to be traced to obtain multiple candidate combination features;

[0162] Determine the source tracing period and divide it into multiple time windows;

[0163] Based on the number of times the first day of occurrence, the time of suddenness, and the independence weight of the candidate combination features within each time window, a comprehensive score for each time window is obtained, and the target time window with the highest comprehensive score is determined.

[0164] Based on the event occurrence time, a secondary screening of candidate combination features within the target time window is performed to obtain the query feature cluster.

[0165] In some optional implementations, the second processing module 502 is further configured to:

[0166] Based on the earliest occurrence date of each candidate combination feature within the target time window, the earliest occurrence date with the most occurrences is determined as the maximum co-occurrence date;

[0167] If the earliest occurrence date of any candidate combination feature is detected to be earlier than the maximum co-occurrence date by at least a certain number of days, the candidate combination feature will be removed from the target time window.

[0168] If the earliest occurrence date of any candidate combination feature is detected to be earlier than the event occurrence time, and the number of occurrences of the first day of the candidate combination feature is higher than the occurrence threshold, the candidate combination feature will be removed from the target time window.

[0169] Based on the remaining candidate combination features within the target time window, the query feature cluster is obtained.

[0170] In some optional implementations, the second processing module 502 is further configured to:

[0171] The event subject and event description are generalized to construct a generalized feature library; the generalized feature library includes multiple generalized subjects and multiple generalized descriptions;

[0172] By combining any generalization subject and generalization description in the generalization feature library, the source-tracing combined features are obtained.

[0173] In some optional implementations, the second processing module 502 is further configured to:

[0174] Cross-validate whether each query feature in a query feature cluster belongs to the same event;

[0175] If outlier query features that do not belong to the same event are identified, retrieve the target Internet information corresponding to the time of the event, and verify the outlier query features based on the target Internet information;

[0176] If the outlier query feature fails the validation, the outlier query feature will be removed from the query feature cluster.

[0177] In some optional implementations, the second processing module 502 is further configured to:

[0178] Retrieve the list of source addresses corresponding to each query feature;

[0179] If overlapping source addresses are detected in the source address lists of any number of query features, then the number of query features is determined to belong to the same event.

[0180] In some optional implementations, the traceability content includes at least some of the following: text content, image content, audio content, and video content; the fourth processing module 504 is further configured to:

[0181] Extract multimodal feature information from the source content;

[0182] Add multimodal feature information to the information to be traced to obtain the updated information to be traced.

[0183] The information tracing device provided in this embodiment of the invention can execute the information tracing method provided in any embodiment of the invention, and has the corresponding functional modules and beneficial effects for executing the method. Further functional descriptions of the various modules and units described above are the same as in the corresponding embodiments described above, and will not be repeated here.

[0184] Figure 6 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention.

[0185] The following is a detailed reference. Figure 6 This diagram illustrates a suitable structural design for implementing an electronic device according to embodiments of the present invention. The electronic device may include a processor (e.g., a central processing unit, graphics processor, etc.) 601, which can perform various appropriate actions and processes based on a program stored in read-only memory (ROM) 602 or a program loaded from memory 608 into random access memory (RAM) 603. RAM 603 also stores various programs and data required for the operation of the electronic device. The processor 601, ROM 602, and RAM 603 are interconnected via a bus 604. An input / output (I / O) interface 605 is also connected to the bus 604.

[0186] Typically, the following devices can be connected to I / O interface 605: input devices 606 including, for example, touchscreens, touchpads, keyboards, mice, cameras, microphones, accelerometers, gyroscopes, etc.; output devices 607 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; memory devices 608 including, for example, magnetic tapes, hard disks, etc.; and communication devices 609. Communication device 609 allows electronic devices to communicate wirelessly or wiredly with other devices to exchange data. Although Figure 6 Electronic devices with various devices are shown, but it should be understood that it is not required to implement or have all of the devices shown, and more or fewer devices may be implemented or have instead.

[0187] In particular, according to embodiments of the present invention, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of the present invention include a computer program product comprising a computer program carried on a non-transitory computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device 609, or installed from a memory 608, or installed from a ROM 602. When the computer program is executed by the processor 601, it performs the functions defined in the information tracing method of the embodiments of the present invention.

[0188] Figure 6 The electronic device shown is merely an example and should not be construed as limiting the functionality and scope of use of the embodiments of the present invention.

[0189] This invention also provides a computer-readable storage medium. The methods described above according to embodiments of the invention can be implemented in hardware or firmware, or implemented as computer code that can be recorded on a storage medium, or implemented as computer code downloaded via a network and originally stored on a remote storage medium or a non-transitory machine-readable storage medium and then stored on a local storage medium. Thus, the methods described herein can be processed by software stored on a storage medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware. The storage medium can be a magnetic disk, optical disk, read-only memory, random access memory, flash memory, hard disk, or solid-state drive, etc.; further, the storage medium can also include combinations of the above types of memory. It is understood that computers, processors, microprocessor controllers, or programmable hardware include storage components capable of storing or receiving software or computer code. When the software or computer code is accessed and executed by the computer, processor, or hardware, the information tracing method shown in the above embodiments is implemented.

[0190] A portion of this invention can be applied as a computer program product, such as computer program instructions, which, when executed by a computer, can invoke or provide the methods and / or technical solutions according to the invention through the operation of the computer. Those skilled in the art will understand that the forms in which computer program instructions exist in a computer-readable medium include, but are not limited to, source files, executable files, installation package files, etc. Correspondingly, the ways in which computer program instructions are executed by a computer include, but are not limited to: the computer directly executing the instructions, or the computer compiling the instructions and then executing the corresponding compiled program, or the computer reading and executing the instructions, or the computer reading and installing the instructions and then executing the corresponding installed program. Here, the computer-readable medium can be any available computer-readable storage medium or communication medium accessible to a computer.

[0191] Although embodiments of the invention have been described in conjunction with the accompanying drawings, those skilled in the art can make various modifications and variations without departing from the spirit and scope of the invention, and such modifications and variations all fall within the scope defined by the appended claims.

Claims

1. An information tracing method, characterized in that, The method includes: Construct a feature decomposition library, which includes multiple combined features and time parameters for each combined feature. The combined features include event subject features and event description features of the corresponding Internet information. Once the information to be traced is determined, the event subject, event description, and event occurrence time of the information to be traced are retrieved from the decomposed feature library to obtain the query feature cluster; Based on the query feature cluster, a source tracing search is performed to obtain the source content, determine the earliest issuance time of the source content, and obtain the source tracing time of the current record. If the earliest sent time is detected to be earlier than the tracing time, the tracing time is updated, and the information to be traced is updated based on the tracing content. The process returns to the step of determining the information to be traced until the earliest sent time is no earlier than the tracing time, and the tracing result is obtained based on the tracing content.

2. The information tracing method according to claim 1, characterized in that, The traceability content includes at least some of the following: text content, image content, audio content, and video content; The step of updating the source information based on the source tracing content includes: Extract the multimodal feature information of the source-tracing content; The multimodal feature information is added to the information to be traced to obtain the updated information to be traced.

3. The information tracing method according to claim 1, characterized in that, The time parameters include the earliest occurrence date, the number of occurrences on the first day, the suddenness time, and the independence weight; based on the event subject, event description, and event occurrence time of the information to be traced, the decomposed feature library is retrieved to obtain the query feature cluster, including: Based on the event subject and event description of the information to be traced, a combination feature to be traced is constructed, and the decomposed feature library is retrieved based on the combination feature to be traced to obtain multiple candidate combination features; Determine the source tracing time period and divide the source tracing time period into multiple time windows; Based on the number of times the first day of occurrence, the time of suddenness, and the independence weight of the candidate combination features within each time window, a comprehensive score for each time window is obtained, and the target time window with the highest comprehensive score is determined. Based on the event occurrence time, a secondary screening is performed on the candidate combination features within the target time window to obtain the query feature cluster.

4. The information tracing method according to claim 3, characterized in that, The process of performing a secondary screening of candidate combination features within the target time window based on the event occurrence time yields a query feature cluster, including: Based on the earliest occurrence date of each candidate combination feature within the target time window, the earliest occurrence date with the most occurrences is determined as the maximum co-occurrence date; If the earliest occurrence date of any candidate combination feature is detected to be earlier than the maximum co-occurrence date by at least a few days, the candidate combination feature is removed from the target time window; If the earliest occurrence date of any candidate combination feature is detected to be earlier than the event occurrence time, and the first occurrence date of the candidate combination feature appears more than the occurrence frequency threshold, the candidate combination feature is removed from the target time window; Based on the remaining candidate combination features within the target time window, a query feature cluster is obtained.

5. The information tracing method according to claim 3, characterized in that, The process of constructing a combined feature for tracing the event based on the event subject and event description of the information to be traced includes: The event subject and the event description are generalized to construct a generalized feature library; wherein, the generalized feature library includes multiple generalized subjects and multiple generalized descriptions; By combining any generalization subject and generalization description in the generalization feature library, the source-tracing combined feature is obtained.

6. The information tracing method according to any one of claims 1-5, characterized in that, The method further includes: Cross-validate whether each query feature in the query feature cluster belongs to the same event; If an outlier query feature that does not belong to the same event is identified, the target Internet information corresponding to the time of the event is retrieved, and the outlier query feature is verified based on the target Internet information. If the outlier query feature fails the validation, the outlier query feature is removed from the query feature cluster.

7. The information tracing method according to claim 6, characterized in that, The cross-validation of whether each query feature in the query feature cluster belongs to the same event includes: Retrieve the list of source addresses corresponding to each query feature; If overlapping source addresses are detected in the source address lists of any number of query features, then it is determined that any number of query features belong to the same event.

8. An information traceability device, characterized in that, The device includes: The first processing module is used to construct a disassembly feature library, which includes multiple combined features and time parameters for each combined feature. The combined features include event subject features and event description features of the corresponding Internet information. The second processing module is used to determine the information to be traced, and based on the event subject, event description and event occurrence time of the information to be traced, to retrieve the disassembly feature library to obtain query feature clusters; The third processing module is used to perform source tracing search based on the query feature cluster, obtain the source content, determine the earliest time of the source content, and obtain the source tracing time of the current record. The fourth processing module is used to update the tracing time if the earliest issuance time is detected to be earlier than the tracing time, and update the information to be traced based on the tracing content, and return to the step of determining the information to be traced, until the earliest issuance time is no earlier than the tracing time, and obtain the tracing result based on the tracing content.

9. An electronic device, characterized in that, include: A memory and a processor are interconnected, the memory stores computer instructions, and the processor executes the information tracing method of any one of claims 1 to 7 by executing the computer instructions.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions for causing the computer to perform the information tracing method according to any one of claims 1 to 7.

Images