Construction method of hot event classification model, hot event classification method and device

Abstract

The present invention provides a method for constructing a hot event classification model, a hot event classification method, and an apparatus for hot event classification. The method utilizes collected event data corresponding to a target hot event to obtain a data sequence, calculates multiple index values ​​for the target hot event based on the data sequence, and calculates a spatial sample of the target hot event based on the multiple index values. The spatial sample is then used to perform cluster analysis on the target hot event to construct a hot event classification model. The hot event classification model classifies the events to be classified to predict their event types, meeting the needs of enterprises for the classification and prediction of real-time public opinion hot events.

Description

Technical Field

[0001] The present invention relates to the field of Internet technology, and in particular to a method for constructing a hot event classification model, a hot event classification method and a device. Background Art

[0002] With the rapid development of information technology, small and medium-sized enterprises serving the public can use public opinion hot events to promote their own products and services, create marketing events, increase product and service exposure, and increase brand awareness. Therefore, enterprises have a growing demand for classified prediction of real-time public opinion hot events.

[0003] Therefore, there is an urgent need for a classification and prediction method for real-time public opinion hot events to meet the classification and prediction needs of enterprises for real-time public opinion hot events. Summary of the Invention

[0004] In view of this, an embodiment of the present invention provides a method for constructing a hot event classification model, a hot event classification method and a device to meet the classification and prediction needs of enterprises for real-time public opinion hot events.

[0005] To achieve the above objectives, the embodiments of the present invention provide the following technical solutions:

[0006] A first aspect of the present invention discloses a method for constructing a hot event classification model, the method comprising:

[0007] Periodically collecting event data of a plurality of hot events at preset time intervals to obtain a data sequence corresponding to each of the hot events, wherein the data sequence includes the event data sorted according to collection time;

[0008] According to the number of collections of each hot event, the hot events whose number of collections is not less than a preset number are selected as target hot events;

[0009] Determine multiple index values ​​of the target hotspot event by using the data sequence corresponding to the target hotspot event;

[0010] Determining a spatial sample corresponding to the target hotspot event based on multiple indicator values ​​of the target hotspot event;

[0011] Clustering all the target hotspot events according to the spatial samples to obtain the final center point coordinates of multiple classification clusters, each of the classification clusters corresponding to an event category;

[0012] Based on the final center point coordinates of the multiple classification clusters, a hot event classification model is constructed.

[0013] Preferably, the event data at least includes collection time, event ranking and ranking weight;

[0014] The determining of multiple index values ​​of the target hotspot event by using the data sequence corresponding to the target hotspot event includes:

[0015] Calculating the overall weight of the target hot event and the initial fluctuation coefficient of the target hot event according to the ranking weights corresponding to the target hot event;

[0016] Calculate the average ranking of the target hot event by using all the event rankings corresponding to the target hot event;

[0017] Calculate the ranking standard deviation of the target hot event based on the ranking mean;

[0018] According to each of the collection times corresponding to the target hotspot event, the duration of the target hotspot event and the number of drop points of the target hotspot event are calculated.

[0019] Preferably, the process of calculating the initial fluctuation coefficient of the target hotspot event includes:

[0020] According to a preset time period, obtaining from the data sequence a first ranking weight corresponding to the first event data within the preset time period, and a second ranking weight corresponding to the last event data within the preset time period;

[0021] The initial fluctuation coefficient of the target hot event is calculated using the first ranking weight and the second ranking weight.

[0022] Preferably, determining the spatial sample corresponding to the target hotspot event based on multiple indicator values ​​of the target hotspot event includes:

[0023] Establishing a spatial vector according to the multiple indicator values ​​of the target hotspot event;

[0024] The spatial vector of the target hotspot event is standardized using a pre-established standardization matrix to determine the spatial sample corresponding to the target hotspot event.

[0025] Preferably, clustering all the target hotspot events according to the spatial samples to obtain the final center point coordinates of multiple classification clusters includes:

[0026] determining a plurality of preset classification values;

[0027] For each of the preset classification values, clustering multiple target hotspot events using the preset classification value and all the spatial samples to obtain initial center point coordinates of multiple classification clusters, and determining a silhouette coefficient corresponding to the preset classification value based on the initial center point coordinates of the multiple classification clusters;

[0028] Compare the silhouette coefficients corresponding to the preset classification values, and determine the preset classification value having the silhouette coefficient closest to 1 as the benchmark classification value;

[0029] The initial center point coordinates of the multiple classification clusters corresponding to the benchmark classification values ​​are recorded as the final center point coordinates.

[0030] A second aspect of an embodiment of the present invention discloses a hot event classification method, which is applicable to a hot event classification model constructed by the method disclosed in the first aspect of an embodiment of the present invention, comprising:

[0031] Determining the hot event classification model;

[0032] Periodically acquiring event data of an event to be classified according to a preset time interval to obtain a data sequence corresponding to the event to be classified;

[0033] Calculating multiple index values ​​of the event to be classified using the data sequence;

[0034] Establishing a spatial sample corresponding to the event to be classified according to the multiple indicator values;

[0035] The spatial samples are input into the hot event classification model for calculation to obtain the event category to which the event to be classified belongs.

[0036] Preferably, the step of inputting the spatial sample into the hot event classification model for calculation to obtain the event category to which the event to be classified belongs includes:

[0037] Calculating the distance between the spatial sample corresponding to the event to be classified and the final center point coordinates of each classification cluster in the hot event classification model;

[0038] Compare the multiple distances obtained by calculation, determine that the classification cluster corresponding to the final center point coordinates with the smallest distance value is the classification cluster to which the event to be classified belongs, and determine that the event category corresponding to the classification cluster to which the event to be classified belongs is the event category of the event to be classified.

[0039] A third aspect of an embodiment of the present invention discloses a device for constructing a hot event classification model, comprising:

[0040] A collection unit, configured to periodically collect event data of a plurality of hot events at preset time intervals to obtain a data sequence corresponding to each of the hot events, wherein the data sequence includes the event data sorted according to collection time;

[0041] A screening unit, configured to screen, based on the number of collections of each hot event, the hot events whose number of collections is not less than a preset number as target hot events;

[0042] A first determining unit, configured to determine multiple indicator values ​​of the target hotspot event by using the data sequence corresponding to the target hotspot event;

[0043] A second determining unit is configured to determine a spatial sample corresponding to the target hotspot event based on multiple indicator values ​​of the target hotspot event;

[0044] A clustering unit, configured to cluster all the target hotspot events according to the spatial samples to obtain the final center point coordinates of a plurality of classification clusters, each of which corresponds to an event category;

[0045] A construction unit is used to construct a hot event classification model based on the final center point coordinates of the multiple classification clusters.

[0046] Preferably, the event data at least includes collection time, event ranking and ranking weight;

[0047] The first determining unit includes:

[0048] A first calculation module is configured to calculate the overall weight of the target hot event and the initial fluctuation coefficient of the target hot event according to the ranking weights corresponding to the target hot event;

[0049] A second calculation module is used to calculate the ranking mean of the target hot event by using all the event rankings corresponding to the target hot event;

[0050] A third calculation module is used to calculate the ranking standard deviation of the target hot event based on the ranking mean;

[0051] The fourth calculation module is used to calculate the duration of the target hotspot event and the number of drop points of the target hotspot event according to each of the collection times corresponding to the target hotspot event.

[0052] A fourth aspect of an embodiment of the present invention discloses a hot event classification device, comprising:

[0053] A determination unit, configured to determine a hot event classification model constructed by the method disclosed in the first aspect of the embodiment of the present invention;

[0054] an acquiring unit, configured to periodically acquire event data of an event to be classified at a preset time interval to obtain a data sequence corresponding to the event to be classified;

[0055] a calculation unit, configured to calculate multiple index values ​​of the event to be classified using the data sequence;

[0056] An establishing unit, configured to establish a spatial sample corresponding to the event to be classified according to the multiple indicator values;

[0057] The classification unit is used to input the spatial sample into the hot event classification model for calculation to obtain the event category to which the event to be classified belongs.

[0058] Based on the above-mentioned embodiment of the present invention, a method for constructing a hot event classification model, a hot event classification method, and a device are provided. The construction method uses the collected event data corresponding to the target hot event to organize the data sequence, calculates multiple index values ​​of the target hot event based on the data sequence, calculates the spatial sample of the target hot event based on the multiple index values, and uses the spatial sample to perform cluster analysis on the target hot event to construct a hot event classification model. The hot event classification model is used to classify the events to be classified to predict the event type of the events to be classified, meeting the enterprise's classification and prediction needs for real-time public opinion hot events. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are merely embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying any creative work.

[0060] Figure 1 A flowchart of a method for constructing a hot event classification model provided by an embodiment of the present invention;

[0061] Figure 2 A flowchart of a hot event classification method provided by an embodiment of the present invention;

[0062] Figure 3 A structural block diagram of a device for constructing a hot event classification model provided by an embodiment of the present invention;

[0063] Figure 4 This is a structural block diagram of a hot event classification device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0064] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0065] In this application, the terms "comprises," "comprising," or any other variations thereof are intended to encompass non-exclusive inclusion, such that a process, method, article, or apparatus that includes a list of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not preclude the presence of additional identical elements in the process, method, article, or apparatus that includes the element.

[0066] As we can see from the background, small and medium-sized enterprises (SMEs) hope to leverage public opinion hotspots to promote their products and services, creating marketing events that increase product and service exposure and brand awareness. Therefore, a classification and prediction method for real-time public opinion hotspots is urgently needed to meet these needs.

[0067] Therefore, embodiments of the present invention provide a method, method, and apparatus for constructing a hot event classification model. These methods construct a hot event classification model based on collected hot event data. Using the hot event classification model, the hot event classification model is used to classify the events to be classified, thereby improving the efficiency of hot event classification.

[0068] join Figure 1 , shows a flow chart of a method for constructing a hot event classification model provided by an embodiment of the present invention, the method for constructing a hot event classification model includes:

[0069] Step S101: periodically collecting event data of multiple hot events at preset time intervals to obtain a data sequence corresponding to each hot event.

[0070] It is understandable that the preset time interval may be Δt, and list data is obtained from the hot event list based on Δt.

[0071] It should be noted that if the refresh period of the hot event list is Δt′, then the refresh period Δt′≥ the preset time interval Δt, that is, the preset time interval Δt needs to be less than the refresh period Δt′ in order to obtain the list data on the hot event list without omission.

[0072] It can be understood that the list data on the hot event list is obtained based on the preset time interval Δt, and the event data of multiple hot events are collected from the list data, wherein the event data includes the event name, collection time, event ranking and ranking weight.

[0073] An event data form is established based on the collected event data. The event data form records the event data of hot events with id, event name (event_name), event ranking (rank), ranking weight (value) and collection time (datetime) as metadata, such as recording the event data of hot events in the form of "(id, event_name, rank, value, datetime)".

[0074] It is understandable that after waiting for a number of periods with a preset time interval of Δt, event data of multiple hot events (events) can be collected, and multiple event data can be collected for each hot event.

[0075] In some specific embodiments, within the data collection time T, N times of list data can be collected from the hot event list according to the preset time interval Δt. The list data collected each time contains the corresponding hot event. N can be called the number of periodic collection times. The specific method for determining N is as shown in formula (1).

[0076]

[0077] For each hot event E, the data sequence W corresponding to each hot event is obtained = {E1…E n}, where n is the total number of hot events E collected from the hot event list within the data collection time T, that is, n hot events E are collected from the hot event list within the data collection time T (collected once every Δt).

[0078] It should be noted that the data sequence includes event data sorted by collection time.

[0079] In the specific implementation of step S101 , event data of multiple hot events are periodically collected from the hot event list at preset time intervals to obtain a data sequence corresponding to each hot event.

[0080] Step S102: Based on the number of collection times of each hot event, select hot events whose number of collection times is not less than a preset number as target hot events.

[0081] It can be understood that the number of collection times of a certain hot event specifically refers to: the number of times the hot event is collected; the preset number n′ can be set in advance according to the situation of the hot event list, where the preset number n′ should meet the following conditions: n′≤collection times N, and assuming the number of times hot event A is collected n≥n′, then P(A)≥0.95, where P(A) is the proportion of all target hot events that need to be collected in all hot events.

[0082] It should be noted that the number of collection times for a hot event can be determined based on the number of event data stored in the data sequence corresponding to the hot event, or in other words, the number of collection times for a hot event = the number of event data in the data sequence of the hot event.

[0083] For example, the data sequence W corresponding to the hot event E stores event data 1, event data 2, event data 3, event data 4, event data 5, and event data 6; the number of collection times for the hot event E is 6.

[0084] It is understandable that, based on the number of collections of each hot event, the hot events with a collection number not less than a preset number are selected as target hot events E', and the data sequence corresponding to the target hot event E' is recorded as W'={E1'…E' n}.

[0085] Step S103: using the data sequence corresponding to the target hotspot event, determine multiple indicator values ​​of the target hotspot event.

[0086] It should be noted that the multiple indicator values ​​include the overall weight (total_value), initial fluctuation coefficient (Beta), ranking mean (rank_mean), ranking standard deviation (rank_std), duration (last_time) and number of drop points (miss_count).

[0087] In the specific implementation of step S103 , multiple index values ​​of the target hotspot event E′ are determined using the event data stored in the data sequence corresponding to the target hotspot event E′.

[0088] It is understandable that the process of determining multiple indicator values ​​of the target hot event E' is as follows:

[0089] According to the ranking weights corresponding to the target hot event E', the overall weight (total_value) of the target hot event E' is calculated. That is, the ranking weights corresponding to the target hot event E' are summed up to calculate the overall weight of the target hot event E' (which can be used as the overall event heat). The specific content of calculating the overall weight is as follows: Formula (2), where E' ivalue is the ranking weight of the i-th event data of the target hot event E', where i is greater than or equal to 1 and less than or equal to n.

[0090] total_value=∑E′ ivalue (2)

[0091] According to the ranking weights corresponding to the target hot event E', the initial fluctuation coefficient (Beta) of the target hot event E' is calculated, that is, according to the preset time period, the first ranking weight corresponding to the first event data in the preset time period and the second ranking weight corresponding to the last event data in the preset time period are obtained from the data sequence; the initial fluctuation coefficient (Beta) of the target hot event E' is calculated using the first ranking weight and the second ranking weight.

[0092] For example, in a preset time period, the start time is t0 and the end time is t1, wherein the relationship between t1 and t0 is as shown in the following formula (3), where n′ is the preset number of times and Δt is the preset time interval.

[0093] t1=t0+n′·Δt (3)

[0094] According to the above preset time period, the first ranking weight E' corresponding to the first event data in the preset time period is obtained from the data sequence t0value , and the second ranking weight E′ corresponding to the last event data in the preset time period t1value , calculate the initial fluctuation coefficient Beta of the target hot event E'. The specific content of calculating the initial fluctuation coefficient is as follows: Formula (4), where E' t0value is the first ranking weight, E′ t1value is the second ranking weight, ∑E′ t0value is the sum of the ranking weights of all target hot events E' at time point t0; ∑E' t1value It is the sum of the ranking weights of all target hot events E' at time point t1.

[0095]

[0096] Using the rankings of all events corresponding to the target hot event E', the ranking mean (rank_mean) of the target hot event E' is calculated. The specific content of calculating the ranking mean of the target hot event E' is as follows: Formula (5). In Formula (5), ∑E' irank is the sum of the event rankings in all event data of the target hot event E', and n is the number of collection times of the target hot event E'.

[0097]

[0098] The ranking standard deviation (rank_std) of the target hot event E' is calculated based on the ranking mean (rank_mean). The specific content of calculating the ranking standard deviation of the target hot event E' is as follows: Formula (6). In Formula (6), E' irank is the ranking standard deviation.

[0099]

[0100] According to each collection time corresponding to the target hotspot event E', the duration (last_time) of the target hotspot event E' is calculated. The specific content of calculating the duration of the target hotspot event E' is as shown in the following formula (7).

[0101] E′ itime =Δt′(n-1) (7)

[0102] According to the collection time corresponding to the target hot event E', the number of missed points (miss_count) of the target hot event E' is calculated, that is, the number of missed points that the target hot event E' appears on the hot event list again after being removed from the hot event list within the period (which can be used as the number of missed points cycles). The specific content of calculating the number of missed points of the target hot event E' is as follows: Formula (8), In formula (8), E' imiss is the number of drop points, E′ iedate is the collection time of the last event data of the target hot event E', E' isdate The time when the first event data of the target hotspot event E' is collected.

[0103]

[0104] Step S104: Based on the multiple indicator values ​​of the target hotspot event, determine the spatial sample corresponding to the target hotspot event.

[0105] It can be understood that the spatial vector A of the target hot event E' is established based on the multiple indicator values ​​of the target hot event E' = {overall weight (total_value), initial fluctuation coefficient (Beta), ranking mean (rank_mean), ranking standard deviation (rank_std), duration (last_time) and number of drop points (miss_count)}.

[0106] It should be noted that the spatial vector of the target hotspot event E' is normalized using a pre-established normalization matrix to determine the spatial sample corresponding to the target hotspot event E'.

[0107] It can be understood that the normalization matrix O is established in advance, and the specific content of the normalization matrix O is as shown in formula (9).

[0108]

[0109] Among them, Robust(), Standard(), and MinMax() are all normalization functions that convert multiple indicator values ​​of the target hot event E'. Specifically, Robust() is a robust normalization function used to standardize indicator values ​​that may have large outliers, such as the overall weight and the number of drop points; Standard() is a variance normalization function used to standardize indicator values ​​such as the ranking mean and duration; MinMax() is an extreme value normalization function used to standardize the ranking standard deviation indicator value with a basically fixed value. For the initial fluctuation coefficient that has already been calculated, there is no need to standardize it again.

[0110] The implementation method of robust normalization is to use information that is robust to outliers to perform normalization, such as taking the range between the first quartile (25th quartile) and the third quartile (75th quartile). The specific implementation method of robust normalization is shown in the following formula (10). In formula (10), X t is robust normalization, X is the index value; is the mean value of the index values ​​of all target hot events; X 0.25 is the first quartile; X 0.75 The third quartile.

[0111]

[0112] The implementation method of extreme value normalization (MinMax) is: use extreme value difference for normalization. The specific implementation method of extreme value normalization (MinMax) is as follows: Formula (11), in Formula (11), X t is the extreme value standardization, X is the index value; X min is the minimum extreme value; X max is the maximum extreme value.

[0113]

[0114] The implementation method of variance standardization (Standard) is: use variance for standardization. The specific implementation method of variance standardization (Standard) is as follows: Formula (12), where X t is the variance standardization, X is the index value; is the mean value of the index of all target hot events; α is the variance.

[0115]

[0116] It can be understood that the spatial sample A′ corresponding to the target hotspot event E′ is calculated using the conversion matrix according to the following formula (13): t .

[0117]

[0118] It should be noted that in formula (13), the transposed matrix A′ of the spatial vector of the target hotspot event E′ is obtained by using the normalized matrix O. T Normalize and get the transposed matrix of the spatial sample corresponding to the target hotspot event E' Then based on the transposed matrix of the spatial samples corresponding to the target hotspot event E' Determine the spatial sample A′ t .

[0119] Step S105: Cluster all target hotspot events according to the spatial samples to obtain the final center point coordinates of multiple classification clusters.

[0120] It can be understood that according to the spatial sample A′ t The specific process of clustering all target hot events E' by using the K-means clustering algorithm is as follows: determine multiple preset classification values ​​K (for example, K = [2, 3, 4, 5, 6]), for each preset classification value K, use the preset classification value K and all spatial samples A' t , cluster multiple target hot events E' to obtain the initial center point coordinates of multiple classification clusters.

[0121] Specifically, from all spatial samples A′ t Select K calibration nodes and calculate the sample space A′ t The distance between each node and each calibration node is recorded as ik; for each node, the node is classified into the calibration node corresponding to the minimum ik of the node to form K classification clusters.

[0122] After obtaining K classification clusters, the silhouette coefficient S corresponding to the preset classification value K is determined based on the initial center point coordinates of the K classification clusters. The specific implementation is as follows: for each classification cluster, calculate the average distance a(i) from the sample space i in the classification cluster to the other sample spaces in the same cluster; and calculate the average distance b(i) from the sample space i in the classification cluster to all sample spaces in other classification clusters, and obtain b(i) = min{bi1, bi2, ..., biK}. Using a(i) and b(i), according to the following formula (14), calculate the silhouette coefficient S corresponding to each preset classification value K. In formula (14), max{a(i), b(i)} is the maximum possible distance between a(i) and b(i) in the sample.

[0123]

[0124] It can be understood that the silhouette coefficients S corresponding to the preset classification values ​​K are compared, and the preset classification value K with the silhouette coefficient S closest to 1 is determined as the benchmark classification value; the initial center point coordinates of multiple classification clusters corresponding to the benchmark classification value are recorded as the final center point coordinates.

[0125] It should be noted that each classification cluster corresponds to an event category.

[0126] It can be understood that all the above distances are Euclidean distances.

[0127] Step S106: constructing a hot event classification model based on the final center point coordinates of multiple classification clusters.

[0128] In the specific implementation of step S106 , a hot event classification model is constructed based on the final center point coordinates of the multiple classification clusters and the event category corresponding to each classification cluster.

[0129] In some embodiments, after constructing a hot event classification model, the event data corresponding to the hot event that has occurred is substituted into the hot event classification model for classification, and the difference between the classification result of the hot event classification model and the actual event category of the hot event that has occurred is determined, and the hot event classification model is corrected according to the difference.

[0130] In an embodiment of the present invention, event data of multiple hot events are used to construct a hot event classification model. Compared with the traditional method of manually identifying hot events and classifying hot events, the embodiment of the present invention does not require manpower costs to classify hot events, saves costs, and improves the efficiency of hot event classification, meeting the company's classification and prediction needs for real-time public opinion hot events.

[0131] Corresponding to the method for constructing a hot event classification model provided by the above embodiment of the present invention, see Figure 2 , an embodiment of the present invention also provides a flowchart of a hot event classification method, comprising the following steps:

[0132] Step S201: Determine a hot event classification model.

[0133] In the specific implementation of step S201 , a hot event classification model is determined. The hot event classification model is constructed based on a method for constructing a hot event classification model disclosed in the above method embodiment.

[0134] Step S202: periodically acquiring event data of the event to be classified at a preset time interval to obtain a data sequence corresponding to the event to be classified.

[0135] In the specific implementation of step S202 , event data of the to-be-classified event that needs to be classified is periodically acquired according to a preset time interval, thereby obtaining a data sequence corresponding to the to-be-classified event.

[0136] Step S203: Calculate multiple index values ​​of the event to be classified using the data sequence.

[0137] In the specific implementation of step S203 , multiple index values ​​of the event to be classified are calculated using the data sequence corresponding to the event to be classified.

[0138] Step S204: establishing spatial samples corresponding to the event to be classified according to multiple indicator values.

[0139] In the specific implementation of step S204 , a spatial sample corresponding to the event to be classified is established based on the calculated multiple indicator values ​​of the event to be classified.

[0140] Step S205: inputting the spatial sample into the hot event classification model for calculation to obtain the event category to which the event to be classified belongs.

[0141] It should be noted that the distance between the spatial sample corresponding to the event to be classified and the final center point coordinates of each classification cluster in the hot event classification model is calculated; the multiple distances obtained by calculation are compared, and the classification cluster corresponding to the final center point coordinates with the smallest distance value is determined to be the classification cluster to which the event to be classified belongs, and the event category corresponding to the classification cluster to which the event to be classified belongs is determined to be the event category of the event to be classified.

[0142] It can be understood that all the above distances are Euclidean distances.

[0143] In the specific implementation of step S205 , the spatial samples corresponding to the event to be classified are input into the hot event classification model, and the event category corresponding to the classification cluster to which the event to be classified belongs is calculated as the event category of the event to be classified.

[0144] In an embodiment of the present invention, a hot event classification model is determined and event data of the event to be classified is obtained, the event data of the event to be classified is calculated to obtain multiple index values ​​of the event to be classified, a spatial sample corresponding to the event to be classified is established based on the multiple index values, the event category to which the event to be classified belongs is calculated through the hot event classification model, and the event to be classified is classified, thereby improving the classification efficiency of hot events.

[0145] Corresponding to the method for constructing a hot event classification model provided by the above embodiment of the present invention, see Figure 3The embodiment of the present invention also provides a structural block diagram of a device for constructing a hot event classification model, which includes: a collection unit 301, a screening unit 302, a first determination unit 303, a second determination unit 304, a clustering unit 305 and a construction unit 306.

[0146] The collection unit 301 is used to periodically collect event data of multiple hot events at preset time intervals to obtain a data sequence corresponding to each hot event, where the data sequence includes event data sorted according to collection time.

[0147] The screening unit 302 is configured to screen, based on the number of collections of each hot event, hot events whose number of collections is not less than a preset number as target hot events.

[0148] The first determining unit 303 is configured to determine multiple indicator values ​​of the target hotspot event by using a data sequence corresponding to the target hotspot event.

[0149] The second determining unit 304 is configured to determine a spatial sample corresponding to the target hotspot event based on multiple indicator values ​​of the target hotspot event.

[0150] In a specific implementation, the second determination unit 304 is specifically used to: establish a spatial vector according to multiple indicator values ​​of the target hotspot event; and standardize the spatial vector of the target hotspot event using a pre-established normalization matrix to determine the spatial sample corresponding to the target hotspot event.

[0151] The clustering unit 305 is used to cluster all target hot events according to the spatial samples to obtain the final center point coordinates of multiple classification clusters, each classification cluster corresponding to an event category.

[0152] The construction unit 306 is used to construct a hot event classification model based on the final center point coordinates of multiple classification clusters.

[0153] In this embodiment of the present invention, the collected event data for multiple hot events is used to calculate multiple indicators for the target hot event. Spatial samples are then generated based on these indicators. The spatial samples are then used to cluster the target hot events and construct a hot event classification model. Using this hot event classification model to classify hot events improves the efficiency of event classification.

[0154] Combine Figure 3 In some embodiments, the event data includes at least the collection time, event ranking, and ranking weight. The first determining unit 303 includes: a first calculation module, a second calculation module, a third calculation module, and a fourth calculation module.

[0155] The first calculation module is used to calculate the overall weight of the target hot event and the initial fluctuation coefficient of the target hot event according to the ranking weights corresponding to the target hot event.

[0156] The second calculation module is used to calculate the ranking mean of the target hot event by using the rankings of all events corresponding to the target hot event.

[0157] The third calculation module is used to calculate the ranking standard deviation of the target hot event based on the ranking mean.

[0158] The fourth calculation module is used to calculate the duration of the target hotspot event and the number of drop points of the target hotspot event according to each collection time corresponding to the target hotspot event.

[0159] In a specific implementation, the process of calculating the initial fluctuation coefficient of the target hot event includes: according to a preset time period, obtaining from the data sequence a first ranking weight corresponding to the first event data in the preset time period, and a second ranking weight corresponding to the last event data in the preset time period; and using the first ranking weight and the second ranking weight to calculate the initial fluctuation coefficient of the target hot event.

[0160] Combine Figure 3 In some embodiments, the clustering unit 305 includes:

[0161] The determination module is used to determine multiple preset classification values.

[0162] The clustering module is used to cluster multiple target hotspot events for each preset classification value using the preset classification value and all spatial samples to obtain the initial center point coordinates of multiple classification clusters, and determine the silhouette coefficient corresponding to the preset classification value based on the initial center point coordinates of multiple classification clusters.

[0163] The comparison module is used to compare the silhouette coefficients corresponding to the preset classification values ​​and determine the preset classification value with the silhouette coefficient closest to 1 as the benchmark classification value.

[0164] The recording module is used to record the initial center point coordinates of multiple classification clusters corresponding to the benchmark classification values ​​as the final center point coordinates.

[0165] Corresponding to a hot event classification method provided by the above embodiment of the present invention, see Figure 4 The embodiment of the present invention also provides a structural block diagram of a hot event classification device, which includes: a determination unit 401, an acquisition unit 402, a calculation unit 403, an establishment unit 404 and a classification unit 405.

[0166] The determining unit 401 is configured to determine a hot event classification model constructed by the method for constructing a hot event classification model disclosed in the above method embodiment.

[0167] The acquiring unit 402 is configured to periodically acquire event data of the event to be classified at a preset time interval to obtain a data sequence corresponding to the event to be classified.

[0168] The calculation unit 403 is used to calculate multiple index values ​​of the event to be classified using the data sequence.

[0169] The establishing unit 404 is configured to establish a spatial sample corresponding to the event to be classified according to multiple indicator values.

[0170] The classification unit 405 is used to input the spatial sample into the hot event classification model for calculation to obtain the event category to which the event to be classified belongs.

[0171] In the embodiment of the present invention, a hot event classification model is determined, and the hot event classification model is used to classify events to be classified, thereby improving the efficiency of hot event classification.

[0172] Preferably, combined Figure 4 , the classification unit 405 includes a calculation module and a determination module:

[0173] The calculation module is used to calculate the distance between the spatial samples corresponding to the event to be classified and the final center point coordinates of each classification cluster in the hot event classification model.

[0174] The determination module is used to compare the multiple distances obtained by calculation, determine that the classification cluster corresponding to the final center point coordinates with the smallest distance value is the classification cluster to which the event to be classified belongs, and determine that the event category corresponding to the classification cluster to which the event to be classified belongs is the event category of the event to be classified.

[0175] To sum up, the embodiments of the present invention provide a method for constructing a hot event classification model, a hot event classification method and a device. Through this construction method, a hot event classification model is constructed, and the hot event classification model is used to classify events to be classified, thereby improving event classification efficiency, reducing the cost of manually classifying hot events, and meeting the company's classification and prediction needs for real-time public opinion hot events.

[0176] Each embodiment in this specification is described in a progressive manner. The same or similar parts between the embodiments can be referred to each other. Each embodiment focuses on the differences from other embodiments. In particular, for system or system embodiments, since they are basically similar to method embodiments, the description is relatively simple. For relevant parts, refer to the partial description of the method embodiment. The system and system embodiments described above are merely schematic, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment. A person of ordinary skill in the art can understand and implement it without expending creative work.

[0177] Professionals may further appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the above description has generally described the components and steps of each example according to their functions. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professionals and technicians may use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the present invention.

[0178] The above description of the disclosed embodiments is intended to enable one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be readily apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the scope of the present invention. Therefore, the present invention is not limited to the embodiments shown herein, but is intended to conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for constructing a hot event classification model, characterized in that: The method comprises: Periodically collecting event data of multiple hot events from a hot event list at a preset time interval to obtain a data sequence corresponding to each of the hot events, wherein the data sequence includes the event data sorted according to the collection time; the preset time interval is less than a refresh period of the hot event list; According to the number of collections of each hot event, the hot events whose number of collections is not less than a preset number are selected as target hot events; Determine multiple index values ​​of the target hotspot event by using the data sequence corresponding to the target hotspot event; Based on multiple indicator values ​​of the target hotspot event, a spatial vector of the target hotspot event is established, and the spatial vector is normalized using a pre-established normalization matrix to determine a spatial sample corresponding to the target hotspot event; Clustering all the target hotspot events according to the spatial samples to obtain the final center point coordinates of multiple classification clusters, each of the classification clusters corresponding to an event category; Based on the final center point coordinates of the multiple classification clusters, a hot event classification model is constructed.

2. The method according to claim 1, characterized in that The event data at least includes collection time, event ranking and ranking weight; The determining of multiple index values ​​of the target hotspot event by using the data sequence corresponding to the target hotspot event includes: Calculate the overall weight of the target hot event and the initial fluctuation coefficient of the target hot event based on the ranking weights corresponding to the target hot event; the initial fluctuation coefficient is calculated by the first ranking weight corresponding to the first event data and the second ranking weight corresponding to the last event data within a preset time period in the data sequence; Calculate the average ranking of the target hot event by using all the event rankings corresponding to the target hot event; Calculate the ranking standard deviation of the target hot event based on the ranking mean; According to each of the collection times corresponding to the target hotspot event, the duration of the target hotspot event and the number of dropouts of the target hotspot event are calculated; the number of dropouts is the number of times the target hotspot event is collected again after not being collected for the first time within a continuous period.

3. The method according to claim 2, characterized in that The process of calculating the initial fluctuation coefficient of the target hotspot event includes: According to a preset time period, obtaining from the data sequence a first ranking weight corresponding to the first event data within the preset time period, and a second ranking weight corresponding to the last event data within the preset time period; The initial fluctuation coefficient of the target hot event is calculated using the first ranking weight and the second ranking weight.

4. The method according to claim 1, wherein The determining of the spatial sample corresponding to the target hotspot event based on the multiple indicator values ​​of the target hotspot event includes: Establishing a spatial vector according to the multiple indicator values ​​of the target hotspot event; The spatial vector of the target hotspot event is standardized using a pre-established standardization matrix to determine the spatial sample corresponding to the target hotspot event.

5. The method according to claim 1, wherein Clustering all the target hotspot events according to the spatial samples to obtain the final center point coordinates of multiple classification clusters includes: determining a plurality of preset classification values; For each of the preset classification values, clustering multiple target hotspot events using the preset classification value and all the spatial samples to obtain initial center point coordinates of multiple classification clusters, and determining a silhouette coefficient corresponding to the preset classification value based on the initial center point coordinates of the multiple classification clusters; Compare the silhouette coefficients corresponding to the preset classification values, and determine the preset classification value having the silhouette coefficient closest to 1 as the benchmark classification value; The initial center point coordinates of the multiple classification clusters corresponding to the benchmark classification values ​​are recorded as the final center point coordinates.

6. A hot event classification method, characterized in that: A hot event classification model suitable for constructing the method according to any one of claims 1 to 5, comprising: Determining the hot event classification model; Periodically acquiring event data of an event to be classified according to a preset time interval to obtain a data sequence corresponding to the event to be classified; Calculating multiple index values ​​of the event to be classified using the data sequence; Establishing a spatial sample corresponding to the event to be classified according to the multiple indicator values; The spatial samples are input into the hot event classification model for calculation to obtain the event category to which the event to be classified belongs.

7. The method according to claim 6, characterized in that The step of inputting the spatial sample into the hot event classification model for calculation to obtain the event category to which the event to be classified belongs includes: Calculating the distance between the spatial sample corresponding to the event to be classified and the final center point coordinates of each classification cluster in the hot event classification model; Compare the multiple distances obtained by calculation, determine that the classification cluster corresponding to the final center point coordinates with the smallest distance value is the classification cluster to which the event to be classified belongs, and determine that the event category corresponding to the classification cluster to which the event to be classified belongs is the event category of the event to be classified.

8. A device for constructing a hot event classification model, characterized in that: include: A collection unit, configured to periodically collect event data of multiple hot events from the hot event list at preset time intervals to obtain a data sequence corresponding to each of the hot events, wherein the data sequence includes the event data sorted by collection time; The preset time interval is shorter than the refresh cycle of the hot event list; A screening unit, configured to screen, based on the number of collections of each hot event, the hot events whose number of collections is not less than a preset number as target hot events; A first determining unit, configured to determine multiple indicator values ​​of the target hotspot event by using the data sequence corresponding to the target hotspot event; A second determining unit is configured to establish a spatial vector of the target hotspot event based on multiple indicator values ​​of the target hotspot event, normalize the spatial vector using a pre-established normalization matrix, and determine a spatial sample corresponding to the target hotspot event; A clustering unit, configured to cluster all the target hotspot events according to the spatial samples to obtain the final center point coordinates of a plurality of classification clusters, each of which corresponds to an event category; A construction unit is used to construct a hot event classification model based on the final center point coordinates of the multiple classification clusters.

9. The device according to claim 8, characterized in that The event data at least includes collection time, event ranking and ranking weight; The first determining unit includes: A first calculation module is configured to calculate the overall weight of the target hot event and the initial fluctuation coefficient of the target hot event based on the ranking weights corresponding to the target hot event; the initial fluctuation coefficient is calculated by the first ranking weight corresponding to the first event data and the second ranking weight corresponding to the last event data within a preset time period in the data sequence; A second calculation module is used to calculate the ranking mean of the target hot event by using all the event rankings corresponding to the target hot event; A third calculation module is used to calculate the ranking standard deviation of the target hot event based on the ranking mean; The fourth calculation module is used to calculate the duration of the target hotspot event and the number of dropouts of the target hotspot event based on the collection times corresponding to the target hotspot event; the number of dropouts is the number of times the target hotspot event is collected again after not being collected for the first time in a continuous period.

10. A hot event classification device, characterized in that: include: A determination unit, configured to determine a hot event classification model constructed by the method according to any one of claims 1 to 5; an acquiring unit, configured to periodically acquire event data of an event to be classified at a preset time interval to obtain a data sequence corresponding to the event to be classified; a calculation unit, configured to calculate multiple index values ​​of the event to be classified using the data sequence; An establishing unit, configured to establish a spatial sample corresponding to the event to be classified according to the multiple indicator values; The classification unit is used to input the spatial sample into the hot event classification model for calculation to obtain the event category to which the event to be classified belongs.

Images