LT6685B - Message sequence analysis - Google Patents

Abstract

The invention subject is the technology for identifying statistical links in the sequence of news items, adverts, or other messages. Incoming messages are classified according to several attributes. Selective reclassification is used to account for different trait assessment interpretations. The messages converted into code form an estimator matrix. To detect a pattern in a message sequence on a timescale, it is necessary to compare matrix fragments which follow either before or after messages with the same assessment value according to one or more traits. The correlation dependence with the same data filter on the superimposed time segments is assessed. If the correlation dependence for two or more matrix fragments is high, the data filter becomes narrower. Data on settings and search results are stored in the database as a pattern. The examples discovered are assessed by a person for significance. A new or repeated pattern search starts with settings combining two or more known patterns with similar message codes. The patterns with high significance assessment are more often used to create combined search settings. The data filter is additionally extended using random values. Figuratively speaking, the pattern search criteria evolve by crossing, mutation, and selection. The analysis predictive power is expressed in the assessment of probability with which the new or probable message fits into the previously identified pattern. The past message sequence examples show what typically happens under similar circumstances. (fig. 10)

Description

The present invention relates to data processing systems and methods specifically adapted for administrative, commercial, financial, managerial or forecasting purposes.

US10181167 describes a method of predicting a politician's actions based on unbound historical data. In this example, there is an obvious link between the subject and the circumstances. Searching for relationships between events with an unspecified number of entities is not foreseen.

A further object of the present invention is to detect statistical relationships in a sequence of news, advertising or other communications. The individual receives an extension of his cognitive abilities through the analysis of collective experience.

The analysis is done by computer. The following system components are required: as a minimum one memory module; as a minimum one processor; as a minimum one data entry interface; as a minimum of one device for displaying or transmitting data. All data mentioned in this description is stored in a memory module. All digital processes are done by the processor.

Messages are classified by several attributes. Symptoms include data about the message and its context with a formal evaluation. In the initial stage, the ratings are given by the individual. The self-paying system uses accumulated material to automate the classification process. In terms of content, the classification is similar to the international patent classification.

The classifier may be treated inaccurately. Estimation deviations are detected by optionally re-classifying the message. What matters to the algorithm is not the cosensity of the application of the estimation, but the repetition of the estimation in different messages. Different ratings for the same message are used to search for similar messages. Increases or decreases in the number of different estimates indicate points of occurrence and disappearance of potential social or individual conflict.

The messages are converted into codes and form a matrix of ratings. The matrix data allow you to plot the following graphs over time: the amount of attention attracted; volume of data; rate of data volume change; other derivative functions. Graphs are formed using data filters to exclude a combination of data of interest. Filters are extraction, inversion, conjuncture, disjunction and the application of coefficients. The ratio can be the number of views to the message or other significance rating. Average correlation values of graphs with different filters are calculated. The local deviation of the mean value reflects the changes specific to a given phenomenon. For example, distinguishable changes are specific to rest day or war period.

The matrix looks for analogues of a single message to detect a duplicate template from multiple messages. A similar message has the same ratings for one or more attributes. Fragments of the matrix that follow or precede analog messages are aligned over time. Correlation between aligned matrix fragments with a uniform data filter is evaluated. If the correlation values are low, the search is repeated with different data filters. Two or more samples with a high correlation value shall be detected. The length of the matrix fragments to be compared is scaled to detect a greater number of samples. Message estimates are sequentially removed from the data filter or inverted to detect the maximum correlation value. The data filter becomes narrower and message codes are identified which ensure high correlation. The search settings and detected examples are stored in the database as a "template".

Multiple templates with similar message codes are used to combine search settings and obtain a broader data filter. The data filter is further expanded using randomly taken values. A broader data filter is used in the new template search procedure.

The analysis can be done against a planned or imagined message. To show the user what usually happens before or after similar circumstances, an example of a message sequence from the past is formed. An interesting example attracts people's attention. People's attention is measured as the number of views or other rating provided. People's attention is the callback to a self-paid algorithm to look for heuristic combinations automatically.

Description of submitted images:

Fig. 1 is an example of a message classifier, index E. Marked positions: 1- Abbreviation for formal evaluation; 2- Definition of Evaluation.

Fig. 2 is an example of a message classifier, index F.

Fig. 3 is an example of a message classifier, index G.

Fig. 4 is an example of a message classifier, index H.

Fig. 5 is an example of a message classifier, index I.

Fig.6 shows an example of a message classifier, indexes J, K.

Fig.7 is an example of a message classifier, index L.

Figure 8 shows an example of converting a message into a code. Marked Positions: 3- message information; 4- message header; 5- code format based on alphanumeric index; 6- message code.

Figure 9 is an example of a matrix. Marked Positions: 7- Attribute Indices; 8- message number; 9- Classified data.

Fig. 10 is a schematic representation of the alignment of matrix fragments. Marked Positions: 10-time scale; 11- message codes; 12- codes for analog messages; 13- aligned fragments of the matrix; 14- estimation of data sequence correlation; 15examples with higher correlation coefficient.

Figure 11 shows a template search procedure. Selected positions: 16code of the message being analyzed; 17- Analyst Passport; 18- message code matrix; 19- Database of templates; 20- Searching for templates with similar codes; 21- the data filter of these templates is removed from the search settings; 22- templates whose data filters are merged; 23- random assessments or attributes; 24- combination of search settings; 25- template search; 26- Detected template data.

Figure 12 shows an example of template data. Selected positions: 27analog filters; 28- number of similar codes in the matrix during the search; 29- data filter at the beginning of the search; 30- length of matrix fragments; 31- data filter at the end of the search; 32- number of detected template samples; 33- numbers of analogous messages; 34- numbers of statistically linked messages; 35- Name given to the template; 36- Materiality rating.

Figure 13 is an example of data volume graphs. Marked Positions: 37- time scale; 38- number of similar messages per year; 39- Number of template samples per year; 40- correlation coefficient of graphs.

Below is an example of analysis of twenty-first century Russian-language news reports. Russia is believed to have been the source of a large number of misleading messages. Therefore, interest in detected templates is also higher.

The top four news reports are taken daily from the news archive. The person converts the messages into code (Fig.8) with the help of a classifier (Figs. 1-7). The message attributes are assigned alphabetical indexes:

A - date of notification;

B - position of the message in the daily news list;

C - heading;

D - link to the message;

E - subject of the message;

F - as described or perceived action;

G - at what stage of execution is the action;

H - time interval between the event and the message;

I - type of information source;

J - place of events or consequences;

K - the place or subject of the cause of the message;

L - assessment of reality from other sources;

M - number and verbal marker of another message about the same phenomenon;

N - circumstances of the notification;

O - calculates message parameters;

P - analyst's passport number.

The number of combinations by indexes E * F * G * H * I * J * K * L represents 1,610,612,736 variants.

Data from other sources clarify the context of the report (index N). In this example, the price of oil is taken as a circumstance.

The message information (index O) is updated each time the algorithm processes the message data. List of estimated message parameters: number of code scans throughout the matrix history; a list of templates where this message is mentioned; human-made assessment of the significance of templates. The list of calculated parameters can be expanded.

The message code (6) may contain an error by the analyst or may be different due to another interpretation. The messages (9, 11) that are already in the matrix are optionally reclassified with a set periodicity. Data on different interpretations are recorded in the analyst's passport (17) as a "paradox". Analyst Passport Data: Human Identifier; number of messages processed; amount of working time; a list of paradoxes; detection dates for paradoxes. If the amount of paradoxes is large, all the work done by the analyst must be verified by another analyst. The list of paradoxes gives variations on the different ratings that are used to search for similar messages (12). This allows the classifier to remain simple and to employ a large number of analysts without specific professional training.

The settings profile (24) is used to search for templates. The setup profile has the following information: length of alignable matrix fragments (30); search direction before or after the message; data filter (29). The first setting profiles are created by a human.

Template detection procedure:

1. Finding similar codes in a matrix (12). An analogous message code has the same rating based on one or more attributes (27).

2. Matrix fragments (13) preceding or following an analogue code are aligned on a time scale (10).

3. Estimation of sequence correlation of matched data (14). One data filter is used for all fragments (13).

4. Samples with higher correlation coefficient (15) are distinguished.

5. The data filter is made narrower (31) to detect codes that provide a high correlation value (15). Message ratings are sequentially inverted or removed. Aligned fragments of the matrix are truncated.

6. The pattern search procedure is repeated with a narrower data filter until the correlation value increases (15) and the number of samples decreases (32-34).

The detected patterns (Figs. 12, 13) are published. The template receives a significance rating from people (36). The number of views, number of comments or other interest rate is evaluated. The analyst can rate the score system (36) from 'not interesting' to 'interesting'. The template may be given a name (35). During publication, templates are sorted by attributes of higher attention. For example, a template example with an existing message rating "denied" (fig.7) attracts more attention. The attributes of higher attention are adjusted based on the statistics of the Template Significance Assessments (36). Templates with higher significance and performance ratings (22) are more commonly used in the message analysis process. Template performance is evaluated by the number of detected samples (32), sample length (30), sample correlation coefficient (15, 40).

The stored template (Fig. 12) has a narrow data filter (30, 31). A combination of settings from multiple templates (24) provides a broader data filter to perform a new template search (25). Procedure for setting profile settings:

1. The message code (16) is comparable to the paradox records (17). If there is more than one variation of estimates, all known variants shall be used.

2. Searching for similar codes in the template database (19). The codes followed by the search (27) and the codes with the detected statistical relationship (31) are read. If the search is repeated for the same message, the previously detected template is not read.

3. Combining data filters from two or more templates with analog codes (22). The number of filters that can be connected is limited (22). The filters can be connected in series or randomly. Templates with high significance rating (36) are used more frequently (22). This increases the likelihood of detecting a more complex and intriguing sample.

4. The data filter removes settings for one or more low-significance templates (21). This reduces the likelihood of a person not being interested in the sample.

5. The length of the matrix fragments to be compared is extended to the time interval between the event and its notification (indexes H, M). This operation may be optional. The choice may be random.

6. Randomized estimates or attributes are used to extend the data filter (23).

In the given procedure, template search settings evolve through combination (22), mutation (23), and selection (20). Each time the matrix (18) is analyzed (25) with a modified filter (24). The message analysis can be repeated many times. Initially, a large number of simple templates are stored in the database (FIG. 12). Later templates with a large number of messages in a single instance appear.

The order of operations of the algorithm, the priority of the data and the thresholds are changed. The number of count operations required to detect template samples is counted. Process optimization is performed with higher performance settings. An example of a threshold value is the correlation coefficient (15) above which a statistical message relationship is considered detected. An example of data priority is the balance between random values (23) and saved template settings (22). The order of operations and the priority of the data are changed by human assistance or by dependence on the calculated values. Cumulative algorithm performance estimates are used to refine the parameters. For example, there is a refinement of the number of templates that are used to form a new profile for the template search settings (22).

The invention is realized as a universal software product. The data entry method and classifier are changed depending on the task.

A first example of using the present invention is a news rating agency. Ranking is the likelihood of a repetition or repetition of a particular message sequence.

A second example of the use of the present invention is the reconstruction of the relationship between the circumstances of human life and the content of advertising addressed to it. The matrix contains data about the content of the advertisement and the recipient of the advertisement. Data is collected from a large number of advertising recipients. Template searches are done on all ad recipients' matrices. The recipient of the advertisement receives detailed information about the detected templates for participating in such investigation.

A third example of the use of the present invention is the study of loyalty circumstances. For example, the detected template represents what distinguishes users with special conditions. Collects messages about conditions from a large number of users.

A fourth example of the use of the present invention is the evaluation of the properties of the presenter. For example, an archive of correspondence from many users is being analyzed. Excludes templates associated with negative feedback. The user receives information on the likely behavior of the reporter's manipulation.

A fifth example of the use of the present invention is the evaluation of the properties of an analyst. For example, human or artificial intelligence is given the task of classifying messages. Different people classify the same messages. Estimates of one person messages form a fragment of a fixed length matrix, the templates represent groups of people with different views.

Claims (13)

Definition A method of analyzing a message sequence for detecting patterns and using formal estimates of message attributes and circumstances to form a value matrix that is stored in one or more memory modules, wherein the processor detects in the array analogs for each message according to one or more attributes; compares, on a time scale, fragments of a matrix that precede or precede analog messages; if a high correlation between matched matrix fragments with a uniform data filter is detected, information about the performed alignment, called a template, is stored in the memory module. 2. The method of analyzing a message sequence according to claim 1, characterized in that the classification of the event message is performed according to the following attributes: date of the message; evaluation of attracted attention; the subject of the report; location of events or consequences; the location or entity of the cause of the event; a description of the action; the stage of completion of the action; the time between the event and the message; type of information source; assessment of reality from other sources; circumstances of notification from other sources of information; communication with other messages about the same event. 3. A method for analyzing a message sequence according to claim 1, characterized in that the message is reclassified; different estimates of the same message are written to the memory module as a paradox; the processor performs a search for message analogs in the matrix using paradox records. 4. The method of analyzing a message sequence according to claim 1, wherein the processor calculates a correlation between aligned matrix fragments using one of the following graphs on a time scale: the amount of attention attracted; volume of data; rate of data volume change; other derivative functions. 5. The method of analyzing a message sequence according to claim 1, characterized in that the processor uses the following data filters: extraction, inversion, conjugation, disjunction, coefficient. 6. The method of analyzing a message sequence according to claim 1, wherein the length of the matrix fragments to be compared includes the time interval between the event and the event message. 7. The method of analyzing a message sequence according to claim 1, wherein the processor shortens matching matrix fragments until the correlation value increases. 8. A method of analyzing a message sequence according to claim 1, wherein the processor optionally removes the data filter or inverts the message estimates and writes a template with a higher correlation coefficient to the memory module. 9. The message sequence analysis method of claim 1, wherein the processor comprises a wider data filter of two or more templates stored in the memory module. A method for analyzing a message sequence according to claim 1 and point 9, characterized in that the human-provided template significance estimate is written to the memory module; the processor is more likely to use templates with a higher significance rating to expand the data filter. 11. A method for analyzing a message sequence according to claim 1, wherein the processor extends the data filter using randomly selected estimates or attributes. 12. A method for analyzing a message sequence according to claim 1, characterized in that the processor changes the order of operations of the algorithm, the data priority and the thresholds; the amount of processor operations for detecting template samples is counted and written to the memory module. 13. A method of analyzing a message sequence according to claim 1, wherein the template search is performed in relation to a scheduled or imaginary message.