CN108345668B - Time sequence matrix thermodynamic diagram visualization method aiming at category comparison - Google Patents

Abstract

The invention relates to a time sequence matrix thermodynamic diagram visualization method aiming at category comparison, and belongs to the technical field of computer graphics and visualization. The method comprises the following implementation steps: selecting a classification dimension, a time dimension and a display dimension from all dimensions of the data set, and classifying the data set according to the classification dimension; mapping time values in the time dimension to time series values; establishing a matrix layout by taking the data classification as rows and the time sequence values as columns, wherein each classification occupies one row and each time sequence value occupies one column; mapping each data record in the data set to a pixel block in a matrix layout, representing data classification by an abscissa, representing time by an ordinate, and mapping display dimension data to colors; for each data classification and each time series value, the average thereof is presented by a bar graph. The invention can comprehensively show the classification characteristics of the data and the time sequence characteristics of each classification, and improve the visual analysis efficiency.

Description

Time series matrix heatmap visualization method for category comparison

technical field

The invention relates to a time series matrix heat map visualization method for category comparison, and belongs to the technical field of computer graphics and visualization.

technical background

In the current data analysis process, classifying data and comparing different classifications is a common form of data analysis, and analyzing the time series change patterns of different classified data is an important prerequisite for data mining and in-depth analysis.

The visualization method has been proved to be an effective means to set parameters before the application of data mining algorithms. It can improve the accuracy and effectiveness of data mining through the mapping and visualization of data and the help of human intelligence.

Heatmaps are often used in visualization methods such as numerical comparisons, but they are still rare in the visualization and analysis of time series data, and the analysis of time series data based on heatmaps combined with matrix layout is even rarer. , the visualization method proposed by the present invention is to display the time series change patterns of different classification data through the time series matrix heat map, and further analyze the comparison between different classification models based on this, so as to analyze the overall time series characteristics of the data set.

SUMMARY OF THE INVENTION

The purpose of the present invention is to propose a time series matrix heat map visualization method for category comparison according to the requirement of time series change data analysis.

The purpose of this invention is to realize through the following technical solutions:

A time series matrix heatmap visualization method for category comparison proposed by the present invention includes the following steps:

Step 1: Select the following types of data dimensions from all dimensions of the dataset:

(1) In the process of classifying all the data in the dataset, the dimension used as the basis for classification (referred to as the classification dimension);

(2) Identify the dimension of the occurrence time of each data record in the data set (referred to as the time dimension);

(3) The dimension used for data analysis in the data set, that is, the dimension used for presentation in the visual analysis process (referred to as the presentation dimension for short).

Step 2: For the classification dimension selected in step 1, all data records in the entire data set are divided into different classifications based on certain classification rules, and the number of classification types is recorded as D.

Step 3: For the time dimension selected in Step 1, map the time value to a time series value, and record the total number of time series values as T.

The time dimension includes month, day, hour, minute, etc., and the time dimension in the data is converted into time series values. The specific conversion method is as follows:

Step 3.1: If the time axis is defined in terms of months and days, the time series value conversion method is shown in formula (1):

DV _i =31×mon _i +d _i (1)

Among them, mon _i and d _i are the month and day corresponding to the i-th data record, respectively, and DV _i is its mapped time series value.

Step 3.2: If the time axis is defined in terms of hours and minutes, the time series value conversion method is shown in formula (2):

TV _i =60× _{hi +m i (} 2)

Among them, h _i and mi are the hours and minutes corresponding to the _i -th data record, respectively, and TV _i is the mapped time series value.

Step 4: For the display dimension selected in step 1, learn the basic distribution of its data, and if necessary, select a reasonable conversion function (such as exponential function, log function) to make the converted data as evenly distributed as possible. For each display dimension, the maximum and minimum values of the data to be displayed are set as Z _max and Z _min .

Step 5: Build the layout structure of the visualization results.

In the layout structure of the visualization result of the present invention, the different categories of all data records are arranged vertically, and in all categories in the data set, each category occupies a row; Series values, one column per time series value.

In this layout, each data record in the data set is mapped to a "pixel block" in the visualization result, and the visual elements corresponding to all data records are presented in the form of a "matrix" according to the time series value and classification corresponding to the record. Referred to in the present invention as "matrix layout". The pixel block rendering area in which all data records are presented in the matrix layout is simply referred to as the "pixel block rendering area".

The calculation method of the number of height pixels occupied by each category in the data set is shown in formula (3):

d=(HH ₁ -H ₂ )/D (3)

Among them, d is the height pixel number occupied by each category; H is the total height pixel number of the set visualization result, which can be set according to the actual situation; H ₁ and H ₂ are the pixels outside the "pixel block drawing area" in the visualization result. The number of height pixels; D is the number of classification categories of the dataset obtained in step 2.

The calculation method of the number of width pixels occupied by each time series value in the data set is shown in formula (4):

t=(WW ₁ -W ₂ )/T (4)

Among them, t is the mapping width of each time series value; W is the total width pixel number of the visualization result, which can be set according to the actual situation; W ₁ and W ₂ are the width pixels outside the "pixel block drawing area" in the visualization result; T is the number of time series values.

Step 6: Set the color map of the dimension data in the data record.

In the present invention, the visualization method of "heat map" is used for reference, and the data to be analyzed is mapped into colors, and colors that can distinguish different values may be used.

Step 7: Calculate the position and color coding of the corresponding "pixel block" in the visualization result for each data record in the dataset. The calculation method of the position and color coding of the pixel block corresponding to a data record is:

Step 7.1: According to the classification obtained in step 2 by the classification dimension of the data record, find the ordinate of the classification in the matrix layout, and set it as its own ordinate.

The ordinates of the pixel blocks corresponding to all data records under the same classification are the same, on the contrary, the ordinates of the pixel blocks corresponding to the data records under different classifications are different;

Step 7.2: According to the time series value obtained in step 3 for the time dimension of the data record, find the abscissa of the category in the matrix layout, and set it as its own abscissa.

The abscissas of the pixel blocks corresponding to all data records of the same time series value are the same, on the contrary, the abscissas of the corresponding pixel blocks corresponding to the data records of different time series values are different;

Step 7.3: According to the display dimension data of the data record, calculate the color coding of the corresponding pixel block as shown in formula (5):

Among them, CLR _p is the color number corresponding to the p-th pixel block, TAM _p is the value of the display dimension data in the data record, and maxTAM and minTAM are the maximum and minimum values of the display dimension data in all data records, respectively.

Step 8: In order to increase the amount of information presented in the final visualization result, the average value of the display dimension data of the "data records corresponding to the pixel blocks of each row" in the matrix layout is mapped to a bar graph (called a categorical dimension bar). Figure), and placed on the left side of the matrix layout, the bar chart can be used to analyze the data comparison of the display dimensions of the data records of different categories.

Step 9: In order to increase the amount of information presented in the final visualization result, the average value of the display dimension data of "the data records corresponding to the pixel blocks of each column" in the matrix layout is mapped to a bar graph (called a time dimension bar). Figure), and placed below the matrix layout, the bar chart can be used to analyze the data comparison of the display dimension of the data records of different time series values.

Step 10: Add classification dimension labels and time dimension labels to the overall layout. Add a category dimension label on the left side of the category dimension bar chart, including the specific name and corresponding information of the category; add a time dimension label below the time dimension bar chart.

So far, the time series matrix heatmap visualization method for category comparison has been completed.

beneficial effect

The time series matrix heatmap visualization method for category comparison proposed in this paper comprehensively displays the time series characteristics of each category of data. Compared with the existing data category comparison analysis methods, this method is more intuitive, comprehensive and easy to use.

Description of drawings

Accompanying drawing 1 is the flow chart of the technical scheme of the present invention;

Accompanying drawing 2 is the principle schematic diagram of the matrix layout method proposed in the present invention;

FIG. 3 is a visualization effect of a time-series matrix layout heatmap when the matrix layout heatmap proposed by the present invention is applied to a movie and TV ratings data set.

Detailed ways

A time series matrix heat map visualization method for category comparison proposed by the present invention, the specific implementation is taking the ratings data of Hunan Satellite TV in 2015 as an example, the number of data records in the data set is 4000, and the data included in the data set The dimensions include: broadcast month, broadcast date, broadcast start hour, broadcast start minute, film and television drama name, total number of episodes, broadcast episode number, and audience rating.

Step 1: Select the following types of data dimensions from all dimensions of the dataset:

(1) The dimension used as the classification basis in the classification process of all the data in the data set (referred to as the classification dimension); in this embodiment, the name of the film and television drama is selected as the classification dimension.

(2) Identify the dimension of the occurrence time of each data record in the data set (referred to as the time dimension); in this embodiment, the broadcast month and the broadcast date are selected as the time dimension.

(3) The dimension used for data analysis in the data set, that is, the dimension used for presentation in the visual analysis process (referred to as the presentation dimension for short). In this embodiment, the audience rating is selected as the presentation dimension.

Step 2: For the classification dimension indicated in Step 1, all data records in the entire data set are divided into different classifications based on certain classification rules, and the number of classification types is recorded as D.

In this embodiment, the data set is divided into 8 categories according to the themes of film and television dramas, and the specific categories are shown in Table 1.

Table 1 Classification of subject categories of film and television dramas

Subject Category Number Subject classification 1 Republic of China 2 military 3 family 4 love 5 youth 6 martial arts 7 Anti-Japanese War 8 column drama

Step 3: For the time dimension indicated in Step 1, map time values to time series values, and record the total number of time series values as T.

In this embodiment, the time dimension includes the broadcast month and broadcast date, so the time axis is defined according to the month and day. For example, the first data record is that a TV drama was broadcast on May 2, then its corresponding mapping is The time series value on the x coordinate is

DV _i =31×mon _i +d _i =31×5+2=157 (6)

Among them, mon _i , d _i are the month and day corresponding to the i-th data record, respectively, and DV _i is the mapped time series value.

Step 4: For the display dimension in Step 1, namely the audience rating, learn the basic distribution of its data, and if necessary, select a reasonable conversion function (such as exponential function, log function) to make the converted data as evenly distributed as possible.

In this embodiment, the maximum value of the audience ratings in the entire data set is Z _max =3.19, and the minimum value is Z _min =0.

Step 5: Build the layout structure of the visualization results.

In the layout structure of the visualization result of the present invention, the different categories of all data records are arranged vertically, that is, the categories of film and television drama themes obtained in step 2 are arranged vertically, and each theme occupies a row; the different time series values of all data records are arranged horizontally. , that is, the time series values of all the movies and TV dramas obtained in step 3 are arranged horizontally, and all the broadcast times in the data set, each time period occupies a column.

In this layout, each data record in the dataset is mapped to a "pixel block" in the visualization result, that is, the ratings of movie and TV dramas are mapped to a "pixel block". The visual elements corresponding to the ratings records of all film and television dramas are presented in the form of a "matrix" according to the time series values and subject classifications corresponding to the records.

In this embodiment, the total number of height pixels of the visualization result is set to 300 pixels; the number of height pixels outside the "pixel block drawing area" in the visualization result, H ₁ =50 pixels, H ₂ =50 pixels; Step 2 in this embodiment The subject matter of the film and television dramas obtained in 2019 is classified into 8 categories. The calculation method of the number of height pixels occupied by each category of film and television subject matter in the dataset is shown in formula (7):

d=(HH ₁ -H ₂ )/D=(300-50-50)/8=25 (7)

Set the total width pixels of the visualization result to 1000 pixels, the width pixels outside the "pixel block drawing area" in the visualization result W ₁ = 50 pixels, W ₂ = 50 pixels, and the number of time series values is 244, then map to The calculation method of the width value of each pixel block on the matrix layout is shown in formula (8):

t=(WW ₁ -W ₂ )/T=(1000-50-50)/244=3.69 (8)

Step 6: Set the color map of the dimension data in the data record.

In the present invention, the visualization method of "heat map" is used for reference, and the data to be analyzed is mapped into colors, and colors that can distinguish different values may be used. In this embodiment, grayscale colors are used, and the RGB values of the selected colors are shown in Table 2.

Table 2 Grayscale color table used by pixel blocks

color number color RGB value 1 (140, 140, 140) 2 (130, 130, 130) 3 (120, 120, 120) 4 (110, 110, 110) 5 (100, 100, 100) 6 (90, 90, 90) 7 (80, 80, 80) 8 (70, 70, 70) 9 (60, 60, 60) 10 (50, 50, 50) 11 (40, 40, 40) 12 (30, 30, 30)

Step 7: Calculate the position and color coding of the corresponding "pixel block" in the visualization result for each data record in the dataset. The calculation method of the position and color coding of the pixel block corresponding to a data record is:

Step 7.1: According to the classification obtained in step 2 by the classification dimension of the data record, find the ordinate of the classification in the matrix layout, and set it as its own ordinate.

Step 7.2: According to the time series value obtained in step 3 for the time dimension of the data record, find the abscissa of the category in the matrix layout, and set it as its own abscissa.

Step 7.3: Calculate the color code of the corresponding pixel block according to the presentation dimension data of the data record.

Taking the first data record broadcast on May 2 as an example, the audience rating is 1.7, then the value of the display dimension data in this data record TAM _p = 1.7, the maximum audience rating maxTAM in all data sets is 3.19, the smallest If the audience rating minTAM is 0, the color coding of the pixel block is obtained as:

That is, the color corresponding to the pixel block is the seventh color in Table 3.

Step 8: In order to increase the amount of information presented in the final visualization result, the average value of the display dimension data of the "data records corresponding to the pixel blocks of each row" in the matrix layout is mapped to a bar graph (called a categorical dimension bar). Figure), and placed on the left side of the matrix layout, the bar chart can be used to analyze the data comparison of the display dimensions of the data records of different categories.

In this embodiment, the average value of all the audience ratings of the same theme category in each row is mapped into a bar graph, which is placed on the left side of the matrix to analyze the comparison of audience ratings under different theme categories.

Step 9: In order to increase the amount of information presented in the final visualization result, the average value of the display dimension data of "the data record corresponding to the pixel block of each column" in the matrix layout is mapped to a bar graph (called a time dimension bar). Figure), and placed below the matrix layout, the bar chart can be used to analyze the data comparison of the display dimension of the data records of different time series values.

In this embodiment, the average audience ratings of each column of the same time-series values are mapped into a bar graph, which is placed below the matrix layout to analyze the comparison of audience ratings on different dates.

Step 10: Add classification dimension labels and time dimension labels to the overall layout. Add a classification dimension label to the left of the classification dimension bar chart, including the specific name and corresponding information of the classification; add a time dimension label below the time dimension bar chart.

In this embodiment, subject categories are added to the left side of the bar graph in the category dimension, and broadcast months are added below the bar graph in the time dimension.

So far, the time series matrix heatmap visualization method for category comparison has been completed.

According to the time sequence matrix heat map visualization effect of the Hunan Satellite TV ratings situation selected in accompanying drawing 3, the conclusions can be drawn include:

(1) For the analysis of the broadcast time of TV dramas, in addition to the long pixel blocks, Hunan Satellite TV also has many shorter pixel blocks, which means that Hunan Satellite TV’s weekly and daily broadcasts coexist, and weekly broadcasts are more common. ;

(2) Based on the analysis of the types of TV dramas broadcast, the average ratings of various types of TV dramas broadcast by Hunan Satellite TV are uneven. lower;

(3) According to the analysis of the broadcast months of TV dramas, Hunan Satellite TV shows the characteristics of high holiday files: the viewing rate of summer files in July, August and September shows a peak trend;

(4) According to the analysis of the broadcast volume of TV dramas, Hunan Satellite TV broadcasts more youth and romance dramas, and youth, love and martial arts dramas are mostly presented in the form of weekly broadcast dramas. The top three categories in terms of ratings are the Republic of China, military and family dramas All are broadcast in the form of daily dramas.

Claims (1)

1. A time sequence matrix thermodynamic diagram visualization method aiming at category comparison comprises the following steps:

step 1: the following data dimensions are selected from all dimensions of the data set:

(1) the classification dimension is the dimension which is used as a classification basis in the process of classifying all data in the data set;

(2) the time dimension is the dimension for identifying the occurrence time of each data record in the data set;

(3) the dimension of presentation, namely the dimension used for data analysis in the data set, and the dimension used for presentation in the visual analysis process;

step 2: aiming at the classification dimensionality selected in the step 1, dividing all data records in the whole data set into different classifications based on a certain classification rule, and recording the number of classification types as D;

and step 3: mapping the time values into time sequence values according to the time dimensions selected in the step 1, and recording the total number of the time sequence values as T;

the time dimension comprises month, day, hour and minute, and the time dimension in the data is subjected to time sequence value conversion, and the specific conversion method comprises the following steps:

step 3.1: if the time axis is defined by month and day, the time series value conversion method is as shown in formula (1):

DV_i＝31×mon_i+d_i(1)

wherein mon_i、d_iRespectively recording the month and day corresponding to the ith data, DV_iA time series value mapped thereto;

step 3.2: if the time axis is defined in terms of hours and minutes, the time-series value conversion method is as shown in formula (2):

TV_i＝60×h_i+m_i(2)

wherein h is_i、m_iRecord corresponding hour and minute, TV, for the ith data_iA time series value mapped thereto;

and 4, step 4: acquiring the basic distribution condition of the data of the display dimensions selected in the step 1, if necessary, selecting a log function or an exponential function for conversion to ensure that the converted data are uniformly distributed as much as possible, and setting the maximum and minimum values of the data to be displayed as Z for each display dimension_max、Z_min；

And 5: the method for establishing the layout structure of the visual result comprises the following specific steps: in the layout structure of the visualization result, different classifications of all data records are longitudinally arranged, and in all classifications in a data set, each classification occupies one line; transversely arranging the different time sequence values of all data records, wherein each time sequence value in a data set occupies one column;

in the layout, each data record in the data set is mapped into a pixel block in a visualization result, visual elements corresponding to all the data records present a matrix form according to a time sequence value and classification corresponding to the record, the matrix layout is called as the matrix layout, and a pixel block drawing area presenting all the data records in the matrix layout is called as the pixel block drawing area for short;

the calculation method of the number of height pixels occupied by each category in the data set is shown in formula (3):

d＝(H-H₁-H₂)/D (3)

wherein d is the number of height pixels occupied by each classification; h is the total height pixel number of the set visualization result, and can be set according to the actual situation; h₁、H₂The height pixel number outside the pixel block drawing area in the visualization result is obtained; d is the number of the classification types of the data set obtained in the step 2;

the calculation method of the number of width pixels occupied by each time-series value in the data set is shown in formula (4):

t＝(W-W₁-W₂)/T (4)

wherein t is the mapping width of each time series value; w is the total width pixel number of the visualization result and can be set according to the actual situation; w₁、W₂The width pixel number outside the pixel block drawing area in the visualization result is obtained; t is the number of time sequence values;

step 6: setting a color mapping table for displaying dimension data in the data record;

mapping data to be analyzed into colors capable of distinguishing different numerical values by using a visualization method of thermodynamic diagram;

and 7: calculating the position and color code of the corresponding pixel block in the visualization result of each data record in the data set, wherein the calculation method of the position and color code of the pixel block corresponding to one data record comprises the following steps:

step 7.1: searching the longitudinal coordinate of the classification in the matrix layout according to the classification obtained in the step 2 by the classification dimension of the data record, and setting the longitudinal coordinate as the longitudinal coordinate of the data record;

the vertical coordinates of the pixel blocks corresponding to all the data records under the same classification are the same, otherwise, the vertical coordinates of the pixel blocks corresponding to the data records under different classifications are different;

step 7.2: searching the horizontal coordinate of the classification in the matrix layout according to the time sequence value obtained in the step 3 according to the time dimension of the data record, and setting the horizontal coordinate as the horizontal coordinate of the classification;

the abscissa of the pixel block corresponding to all the data records with the same time sequence value is the same, whereas the abscissa of the pixel block corresponding to the data records with different time sequence values is different;

step 7.3: according to the presentation dimension data of the data record, the color coding of the corresponding pixel block is calculated as shown in formula (5):

wherein, CLR_pNumbering the color corresponding to the p-th pixel block, TAM_pFor the value of the display dimension data in the data record, maxTAM and minTAM are respectively the maximum value and the minimum value of the display dimension data in all the data records;

and 8: in order to increase the amount of information presented in the final visualization result, the average value of the display dimension data of the data records corresponding to the pixel blocks in each row in the matrix layout is mapped into a bar graph and is placed on the left side of the matrix layout, and the bar graph can be used for analyzing the data comparison condition of the display dimension of the data records of different classifications;

and step 9: in order to increase the amount of information presented in the final visualization result, the average value of the presentation dimension data of the data records corresponding to the pixel blocks in each column in the matrix layout is mapped into a bar graph and placed below the matrix layout, and the bar graph can be used for analyzing the data comparison condition of the presentation dimension of the data records with different time sequence values;

step 10: adding a classification dimension label and a time dimension label to the overall layout, and adding a classification dimension label on the left side of the matrix layout, wherein the classification dimension label comprises a specific name and corresponding information of the classification; time dimension labels are added below the matrix layout.

Images