News text-oriented multi-mode RGB intelligent color matching method and system

Abstract

The invention discloses a news text-oriented multi-mode RGB intelligent color matching method and system, and the method comprises the steps: obtaining news data, calculating a color R channel according to a news type, calculating a color G channel according to a news emotion tendency, and calculating a key word B channel; and aggregating the mean values of the three channels of all news to obtain a news color, and performing secondary aggregation on the news color and a specified color in the projection scene to obtain a target color. According to the method, news types, emotional tendencies and keyword weights are mapped to three channels respectively, and multi-modal representation of'color, namely data 'is formed. A user can quickly capture multi-dimensional information through colors, so that switching and comparison among a plurality of charts are avoided; moreover, the target color required by projection is obtained through secondary aggregation, so that the target color can be matched with the theme atmosphere of the projection scene, the current or local news dynamic can be reflected, background colors or ornaments do not need to be frequently replaced, and the complexity of the layout cost of the activity scene is reduced to a certain extent.

Description

Technical Field

[0001] The present invention belongs to the technical field of news data analysis and visualization, and in particular relates to a multimodal RGB intelligent color matching method and system for news texts. Background Art

[0002] Traditional analysis methods for news data analysis have numerous limitations. Firstly, they often rely on single-dimensional text classification or sentiment analysis, which makes it difficult to comprehensively and intuitively reflect the multi-attribute characteristics of news. Important information, such as news type and sentiment, cannot be clearly presented simultaneously. For example, relying solely on text classification only reveals the category of a news item, but fails to understand its sentiment and the importance of its key content. Similarly, simple sentiment analysis can only capture the sentiment of news items, but fails to capture differences in news types or the distribution of key information. Secondly, existing static color mapping or simple keyword statistics lack dynamic aggregation capabilities. These methods cannot visualize the temporal and spatial variations of news data in real time, failing to meet users' needs for dynamic analysis of news data. In today's rapidly changing information age, the timeliness and dynamic nature of news data are crucial. However, existing technologies are unable to effectively and timely reflect these changes, making it difficult for users to quickly grasp news trends. Furthermore, traditional public opinion analysis systems rely on complex charts or reports to present data. These methods not only have low efficiency in information transmission, but also make users easily confused when browsing a large number of complex charts and reports, and it is difficult to quickly extract key information from them; and when faced with hot events, traditional systems find it difficult to quickly identify their emotional tendencies and the relevance of themes, and are unable to provide timely and effective support for decision-making. For example, in the public opinion caused by an emergency, traditional systems are unable to quickly judge the public's emotional attitudes towards the event and its connection with other related topics, resulting in delayed decision-making. For example, the prior art Chinese patent application CN201811325806.2 discloses a method for designing a mood board interface based on user emotions, but this prior art also relies on simple sentiment analysis and color matching based on the emotional tendencies of the text, which cannot reflect the type differences of news and the distribution of key information. Summary of the Invention

[0003] The purpose of the present invention is to provide a multimodal RGB intelligent color matching method and system for news text, which can partially solve or alleviate the above-mentioned deficiencies in the prior art and solve the problem of information fragmentation in traditional text color matching methods.

[0004] In order to solve the above-mentioned technical problems, the present invention specifically adopts the following technical solutions:

[0005] The first aspect of the present invention is to provide a multimodal RGB intelligent color matching method for news texts, comprising: obtaining news data, and calculating the corresponding color R channel of each news item according to the news type in the news data; calculating the corresponding color G channel of each news item according to the news sentiment tendency in the news data; calculating the corresponding color B channel of each news item based on the keyword weight of each news item; aggregating the mean value of the color R channel, the mean value of the color G channel, and the mean value of the color B channel of all news items to obtain the corresponding news color RGB. 聚合 The RGB value of a designated color reflecting the theme atmosphere in the current projection scene is obtained, and the designated color and the aggregated news color are subjected to secondary aggregation to obtain a target color for projection.

[0006] As an improvement, the steps of calculating the color R channel based on the news type in the news data include: sorting the n news types based on the pre-set importance of each news type to obtain a sequence number corresponding to each news type; wherein, the greater the importance, the larger the corresponding sequence number; removing stop words from each news item in the news data and extracting features; predicting the news type of the news item based on the features; and using the formula: Calculate the color R channel value; where R is the color R channel value, k is the serial number of the type to which the news belongs, n is the number of news types, p is the confidence level, and 256 is the total number of R channels. As an improvement, the steps of calculating the color G channel based on the news sentiment in the news data include: performing Chinese word segmentation on each news item in the news data and converting it into a vector; predicting the sentiment of the news item based on the vector; and using the formula: Calculate the color G channel value; where G is the color G channel value, q is the positive probability, ω1 is the positive weight, ω2 is the negative weight, and 256 is the total number of G channels. As an improvement, the steps for calculating the color B channel include: after segmenting each news item in the news data and marking the part of speech, constructing a co-occurrence matrix; extracting keyword weights based on the contribution matrix; using the formula: Calculate the color B channel value; where B is the color B channel value, ω3 is the verb weight, ω4 is the noun weight, ω5 is the remaining part-of-speech weight, and m is the total number of words. As an improvement, calculate the color R channel, color G channel, and color B channel values ​​corresponding to each news item. Specifically, obtain the correlation between each news item and the theme atmosphere; if the correlation is high, use the formula: Calculate the corresponding color R channel value; use the formula: Calculate the corresponding color G channel value; use the formula: Calculate the corresponding color B channel value; if it is low correlation, use the formula: Calculate the corresponding color R channel value; use the formula: Calculate the corresponding color G channel value; use the formula: Or B=c1*256X to calculate the corresponding color B channel value; wherein a1 and a2, b1 and b2, c1 and c2 are preset weighting coefficients, and a1>a2, b1>b2, c1>c2.

[0007] As a further improvement, the multimodal RGB intelligent color matching method for news text also includes the steps of: calculating the second Euclidean distance between the color obtained by the secondary aggregation and the specified color; judging whether the second Euclidean distance is within a preset Euclidean distance threshold range, and if so, using the color after the secondary aggregation as the target color; otherwise, calculating the first Euclidean distance between the news color obtained by the first aggregation and the specified color, and judging whether the first Euclidean distance between the news color obtained by the first aggregation and the specified color is within a preset Euclidean distance threshold range, and if not, judging whether the first Euclidean distance is greater than or equal to the preset distance threshold, and if so, performing segmented aggregation.

[0008] The second aspect of the present invention is to provide a multimodal RGB intelligent color matching system for news texts, including: a news data capture module for acquiring news data; an RGB value calculation module for calculating the color R channel according to the news type in the news data; calculating the color G channel according to the news emotional tendency in the news data; calculating the color B channel corresponding to each news based on the keyword weight of each news; a color aggregation module for aggregating the mean of the color R channel, the mean of the color G channel, and the mean of the color B channel of all news to obtain the news color; a secondary aggregation module for obtaining the RGB value of the specified color reflecting the theme atmosphere in the current projection scene, and performing secondary aggregation on the specified color and the aggregated news color to obtain the target color for projection.

[0009] A third aspect of the present invention is to provide an RGB intelligent color matching method based on news text, characterized by comprising: obtaining news data and attribute features thereof; the attribute features comprising: news type, or news sentiment, or keyword weight; obtaining a mapping relationship template between news attribute features and three primary colors pre-set by a user; the mapping relationship template comprising: news type mapped as a first primary color, sentiment mapped as a second primary color, and keyword weight mapped as a third primary color; wherein the first primary color, the second primary color, and the third primary color are respectively one of R color, G color, and B color; calculating the color channel values ​​of each of the three primary color channels based on the mapping relationship template and the news attribute features; if the first primary color is R color, the second primary color is G color, and the third primary color is B color, the steps of calculating the color channel values ​​of each of the three primary color channels specifically comprise: calculating the color R channel according to the news type in the news data; calculating the color G channel according to the news sentiment in the news data; calculating the color B channel according to the keyword weight; aggregating the mean value of the color R channel, the mean value of the color G channel, and the mean value of the color B channel of all news to obtain the corresponding news color;

[0010] The steps of aggregating the mean of the color R channel, the mean of the color G channel, and the mean of the color B channel of all news to obtain the news color are as follows: Calculate news color; RGB 聚合 is the RGB value of all news colors after aggregation, R i is the color R channel value of the i-th news, and ∑R i The total value of the R channel of all news colors, G i is the color G channel value of the i-th news, and ∑G i is the total value of the G channel of all news, B i is the color G channel value of the i-th news, ∑B i is the total value of the color B channel of all news, and N is the total number of news.

[0011] Furthermore, the method further includes the steps of obtaining the RGB value of a designated color reflecting the theme atmosphere in the current projection scene, and performing secondary aggregation on the designated color and the aggregated news color to obtain a target color for projection: Wherein, β is the aggregation factor between the specified color and the aggregated color, and β=0-1.

[0012] Furthermore, when the user is a new user, the step of obtaining the mapping relationship between the news attribute features preset by the user and the three primary colors specifically includes the steps of: displaying six preset mapping relationship templates; among which, the most frequently used one is the baseline reference template; when the user arbitrarily specifies a template, and the specified template is not the baseline reference template, respectively generating corresponding example news colors based on the baseline reference template and the template selected by the user, as well as all news within a preset historical time period; wherein, the preset historical time period is the most recent week or the most recent month; when the user selects an example news color, its corresponding template is used as the template specified by the user to obtain the mapping relationship between the news attribute features and the three primary colors.

[0013] Furthermore, if the user is an old user, the step of obtaining the mapping relationship between the news attribute features preset by the user and the three primary colors specifically includes the following steps: calculating the preference index Ti = wi*(Ni / M) of each template based on the frequency of use of each template by the old user, wherein Ti is the preference index of the i-th template, Ni is the number of times the i-th template is used by the old user, and M is the total number of times all templates are used by the old user; wi is the weight of each template; displaying at least one template with a preference index greater than a preset threshold and a benchmark reference template to the old user; the benchmark reference template is the template with the highest frequency of use among the six preset mapping relationship templates; when the user arbitrarily specifies a template, and the specified template is not the benchmark reference template, generating corresponding example news colors based on the benchmark reference template and the template selected by the user, as well as all news within a preset historical time period; wherein the preset historical time period is the most recent week or the most recent month; when the user selects an example news color, its corresponding template is used as the template specified by the user to obtain the mapping relationship between the news attribute features and the three primary colors.

[0014] The fourth aspect of the present invention is to provide an RGB intelligent color matching system based on news text, which includes: a first data acquisition module for acquiring news data and its attribute characteristics; the attribute characteristics include: news type, or news emotional tendency, or keyword weight; a second data acquisition module for acquiring a mapping relationship template between news attribute characteristics and three primary colors pre-set by a user; the mapping relationship template includes: news type is mapped to the first primary color, emotional tendency is mapped to the second primary color, and keyword weight is mapped to the third primary color; wherein the first primary color, the second primary color and the third primary color are respectively one of R color, G color and B color; an RGB value calculation module is used to calculate the color channel value of each of the three primary color channels based on the mapping relationship template and the news attribute characteristics; a color aggregation module is used to aggregate the mean value of the color R channel, the mean value of the color G channel, and the mean value of the color B channel of all news in a certain time period into the news color of the time period. As an improvement, the RGB intelligent color matching system based on news text also includes: a secondary aggregation module for obtaining the RGB value of the specified color that reflects the theme atmosphere in the current projection scene, and performing secondary aggregation on the specified color and the aggregated news color to obtain the target color for projection. Furthermore, the RGB intelligent color matching system based on news text also includes: a user type identification module for identifying the type of the current user; the types include new users and old users; an interaction module for displaying six preset mapping relationship templates to the current user when it is identified as a new user, and when the current user arbitrarily specifies a template, and the specified template is not the benchmark reference template, generating corresponding example news colors based on the benchmark reference template and the template selected by the user, as well as all news within the preset historical time period; wherein the benchmark reference template is the most frequently used one of the six templates; the preset historical time period is the most recent week or The last month; a template designation module, which is used to use the corresponding template as the user-designated template when the current user selects an example news color, and obtain the mapping relationship between news attribute features and the three primary colors; a template recommendation module, which is used to calculate the preference index Ti=Ni / M of each template based on the frequency of each template being used by the old user when the current user is identified as an old user, and trigger the interaction module to display at least one template with a preference index greater than a preset threshold and a benchmark reference template to the old user; wherein Ti is the preference index of the i-th template, Ni is the number of times the i-th template is used by the old user, and M is the total number of times all templates are used by the old user. Furthermore, the RGB intelligent color matching system based on news text also includes: a secondary aggregation module, which is used to obtain the RGB value of the specified color reflecting the theme atmosphere in the current projection scene, and perform secondary aggregation on the specified color and the aggregated news color to obtain the target color for projection.

[0015] Beneficial effects: 1. Traditional text color schemes can only display a single dimension. For example, news classification results are marked with fixed colors, which cannot simultaneously present emotional tendencies and keyword weights, resulting in information fragmentation. Alternatively, news emotions are marked with specific colors, which cannot simultaneously present different news types and key information. The present invention maps news types, emotional probabilities, and keyword weights to three channels of RGB colors, forming a multimodal representation of "color as data". Users can quickly capture multidimensional information through color, avoid switching and comparing between multiple charts, and improve data analysis efficiency. 2. Traditional text color schemes are static color mappings that cannot reflect the uncertainty or dynamic characteristics of the data. The color values ​​of the present invention are directly calculated by the model prediction results, rather than a preset mapping table. The color space (256×256×256≈16 million colors) is much larger than the fixed color table of the traditional scheme (usually only dozens of colors), which avoids color conflicts and accurately represents data differences. Third, when color projection is required in a specific scene (for example, by projecting color on a designated sculpture in an art exhibition), the color obtained by aggregating all news within a specified time period or a specified area can be directly used for projection. In order to improve the degree of fit or matching between the color and the theme color required for the scene (that is, the Euclidean distance between the two must be within a preset threshold), on the one hand, the weight of news types associated with the theme atmosphere in the specific scene (for example, environmental protection, health, etc.) is increased, thereby strengthening the color of such news (while also reducing the effect of mean dilution to a certain extent). On the other hand, the news color obtained by aggregating the mean of each channel is secondary fused with the color specified based on the theme atmosphere in the specific scene, so that the final aggregated color not only meets the scene requirements but also can present current or local news dynamics, etc., and through dynamic projection, there is no need for manual color adjustment or constant replacement of different backgrounds or decorations with static colors. Fourth, traditional text color schemes lack the ability to aggregate in the time dimension, or only perform simple numerical accumulation, and cannot use visualization to show the evolution of data over time. The present invention supports time windows such as hours, days, and weeks, and generates the aggregate color of all news in the time window and the aggregate color of the specified type of news in the time window by average calculation (as shown in Figures 7 and Figure 8As shown, news colors are mapped based on news categories within different time periods, or even the aggregated colors of each news item. Furthermore, the changes in color over time directly reveal temporal and spatial patterns, eliminating the need for manual analysis of reports and reducing interpretation costs. (For example, traditional line chart or bar chart visualizations require interpreting the most common news category from the chart, and then using other charts to obtain information such as sentiment, which is costly.) This allows readers to quickly determine whether the type of news that month has shifted to a higher category, with decreased sentiment and increased keyword weight. Traditional news visualizations typically extract single-dimensional data from massive amounts of news data, such as the sentiment of news or the public opinion surrounding a hot topic. Therefore, single-dimensional data is typically extracted for analysis and visualized using charts or graphs. The resulting visualizations often take up considerable space. For example, a distribution chart or trend chart of news categories for each month of the year can be used. Another example is the change in sentiment within a specific news category. However, for scenarios requiring comprehensive and global analysis, this single-dimensional and fragmented presentation fails to directly capture the desired key information and introduces excessive noise. For example, in order to effectively guide online public opinion, it is necessary to quickly know which type of news is dominant in a certain time period and what the emotional tendency of this type of news is, without having to see other information. Only then can appropriate measures be taken to continuously strengthen the positive guidance of online public opinion, thereby forming a systematic, scientific, and effective online public opinion guidance mechanism. If traditional visualization methods are used, not only is it necessary to map the proportion of each type of news in a year into a bar chart or pie chart, but each type is distinguished by a different color. However, this visualization method cannot simultaneously present information such as the emotional tendency and keywords of this type of news in a year. Therefore, it is necessary to combine charts such as sentiment analysis and keyword analysis to present it. This not only introduces too much interference information, but also requires users to switch between different charts to obtain the desired information. In the present invention, by directly fusing all types of news, emotional tendencies, and keywords in a year into a single overall news color, news analysts can directly obtain the corresponding dominant information from the fused news color. This also converts massive news data into intuitive annual insights, helping to quickly identify social concerns and emotional fluctuations. For example, if the final fusion color of all news within a certain time period is cyan (R12, G120, B60), then the word "war" is likely related to real-time politics, strong negativity, or high frequency. Furthermore, for some application scenarios, such as real-time news streams, the color of each news item is calculated and then dynamically displayed by color in a news stream, allowing users to distinguish different news types and emotions at a glance. It can even detect sudden changes in sentiment within the same news category.6. By presetting multiple mapping templates, users can map different news attributes to different templates according to actual needs, realizing user-personalized news mapping. Different industries can then adopt different mapping templates according to their actual needs to obtain different news colors, greatly improving the user experience. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following is a brief introduction to the drawings required for the embodiments or the description of the prior art. In all drawings, similar elements or parts are generally identified by similar reference numerals. In the drawings, the various elements or parts are not necessarily drawn according to the actual scale. Obviously, the drawings described below are some embodiments of the present invention. For those of ordinary skill in the art, other drawings can also be obtained based on these drawings without inventive work.

[0017] Figure 1a This is a flow chart of Example 1; Figure 1b This is a flow chart of Example 2; Figure 2 This is a functional module diagram of Example 3;

[0018] Figure 3 This is a flow chart of Example 4; Figure 4 This is a flow chart of Example 5; Figure 5 This is a flow chart of Example 6; Figure 6a-6f Schematic diagrams of six preset mapping relationship templates; Figure 7a and Figure 7b An example graph of news colors of different types of news obtained by calculating all news on January 22, 2025, based on the method of Example 1, and mapping the attribute features of the news to the mean of three primary colors and then aggregating them; Figure 8 This is an example of the news colors obtained by calculating the three single primary color means of all current political news on April 20, 2025, based on the method of Example 1, and after aggregation. DETAILED DESCRIPTION

[0019] To make the purpose, technical solutions and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments.

[0020] Example 1: Figure 1aAs shown in the figure, this embodiment provides a multi-modal RGB intelligent color matching method for news texts, which is applied to rapid public opinion analysis. Its specific steps include: S1 Obtain news data. Specifically, in this embodiment, news data is crawled from news platforms through crawler technology and stored in a MySQL database; the crawled fields include id, title, content, time, and url. In this embodiment, the crawler will visit the web pages of eight major news platforms such as People's Daily Online and Xinhua News Agency according to certain rules, locate the news-related information and perform crawling. In addition, the fields crawled and stored in this embodiment include id, title, content, time, and url, and each field has its specific meaning and function: id: The unique identifier of the news, which is used to uniquely identify a news record in the database. It can be an automatically generated integer number, which is convenient for subsequent querying, updating, and deleting of news. title: The title of the news, which usually briefly summarizes the main content of the news. The title is an important basis for users to quickly understand the news theme and may also be used for keyword extraction and classification in subsequent analysis. content: The specific content of the news, which contains detailed information and descriptions. This is the core data for news type classification, sentiment analysis, etc. time: The release time of the news, which records the timeliness of the news. Through the time information, the release trend of the news and the hotspots in different time periods can be analyzed. url: The link address of the news, which points to the web page of the news on the original platform. Through the URL, the original news page can be conveniently accessed to obtain more relevant information, and it can also be used for data verification and update. Data collection and storage provide the basic data for subsequent operations. By collecting data from multiple news platforms and storing it in a structured database, the data can be conveniently managed and processed, providing strong support for subsequent analysis work. At the same time, reasonable field design also helps to improve the availability of data and the accuracy of analysis.

[0021] S2 Calculate the R channel of the color according to the news type in the news data; calculate the G channel of the color according to the news sentiment tendency in the news data; calculate the B channel of the color.

[0022] The steps for calculating the R channel of the color according to the news type in the news data in this embodiment include:

[0023] S211 Perform stop word removal processing on each news in the news data and extract features. Stop words refer to those common words that have no practical meaning for text semantic understanding and analysis, such as "de", "le", "zai" in Chinese, and "the", "and", "is" in English, etc. Removing these stop words can reduce data noise, improve the efficiency and accuracy of subsequent analysis, and make the model focus more on key information. In this embodiment, TF-IDF is used for feature extraction.

[0024] S212 predicts the news type and confidence level of the news item based on the features. Specifically, in this embodiment, an SVM classification model is used to predict which of the 21 preset news types each news item belongs to. In this solution, a trained SVM model is used to classify news text after TF-IDF feature extraction to predict which of the 21 news types it belongs to, such as "Current Affairs," "Economy," or "Sports," providing a classification basis for subsequent color mapping. Of course, other models can also be used to predict news types, such as the BERT model, which is not limited in the present invention.

[0025] S213 uses the formula: Calculate the color R channel value; where R is the color R channel value, k is the serial number of the news type, n is the number of news types, p is the confidence level (obtained at the same time when performing news type identification), and 256 is the total number of R channels. Suppose there is a piece of news. After the previous news type classification step, the SVM model predicts that it belongs to the "1: Current Affairs" category of news, that is, k = 1, and the model's confidence level in this prediction is p = 0.9. Then, according to the R channel value calculation formula, the result is approximately 1.219. In actual applications, since the RGB channel values ​​are integers, the results will be rounded, and here R = 1 after rounding. This means that when the news is visually displayed, the color R channel value is 1. In this way, the news type information is converted into color information, which is convenient for subsequent visual display and analysis. In this embodiment, it is necessary to sort each news type in advance to obtain the serial number corresponding to each news type. The ranking can be set by experts. For example, compared with other types of news, real-time political news usually has the highest priority, so it is ranked at the front, so its serial number is 1, while entertainment news has the lowest priority, so it is ranked last and has the largest serial number.

[0026] In other embodiments, the ranking can be performed based on the authority of the source of each news (for example, from mainstream media, or from self-media, and accordingly, the authority of news from mainstream media is greater than that from self-media), the number of times each news is reprinted (preferably, the sum of the number of times each news of this type is reprinted in all news within a specified time period or a specified area), and the coverage. For example, the importance of each news is calculated as 0.4*α+0.3*β+0.3*γ, where if it comes from mainstream media, the source weight α=1.0, and if it comes from self-media, α=0.3-0.6; β is the number of times the news is reprinted; γ is the total number of regions covered when the news is reprinted. Then the importance of each news is calculated (that is, the sum of the importance of all news of this type in N news items), and sorted to obtain the serial number k of each news type.

[0027] S22 The step of calculating the color G channel according to the emotional tendency of the news in the news data includes: S221 Chinese word segmentation processing is performed on each news in the news data and converted into a vector. Use a word segmentation tool (such as jieba) to split the news text into individual words, for example, "Central Bank cuts interest rates to stimulate the economy" is split into "Central Bank / interest rate cuts / stimulus / economy". Calculate the TF-IDF value of each word and generate a feature matrix. TF (word frequency) reflects the importance of a word in the current text, and IDF (inverse document frequency) reflects the rarity of a word in all texts. The combination of the two highlights key emotional vocabulary, such as "good" and "risk". S222 Predict the emotional tendency of the news based on the vector. Specifically, in this embodiment, the TF-IDF feature matrix is ​​input into the Logistic Regression model, and the positive emotional probability q∈[0,1] is output. q≥0.5 is determined to be positive emotion, and q<0.5 is determined to be negative emotion. Of course, models such as LSTM or Transformer can also be used to predict emotional tendency, and this is not limited in the present invention. S223 uses the formula: Calculate the color G channel value; where G is the color G channel value, q is the positive probability, ω1 is the positive weight, ω2 is the negative weight, and 256 is the total number of G channels. Suppose there is a news article whose sentiment prediction step predicts a positive probability of 0.8. Also, let ω1 = 2 and ω2 = 1. The result calculated using the G channel value calculation formula is approximately 230.4. In practice, since RGB channel values ​​are integers, the result is rounded. Here, G = 230 after rounding.

[0028] The step of calculating the color B channel in S23 includes: S231 segmenting and tagging each news item in the news data, and then constructing a co-occurrence matrix. Segmentation is to split the continuous text into meaningful words, while part-of-speech tagging is to tag each word with its part of speech, such as noun (n), verb (v), adverb (adv), etc. In Chinese processing, commonly used tools such as Jieba can complete this task. Through part-of-speech tagging, the parts of speech of interest, such as nouns, verbs and adverbs, can be filtered out, because words of these parts of speech can often better express the core information of the text. The co-occurrence matrix is ​​used to describe the association between words. In a given text window (for example, a sentence or a paragraph), if two words appear at the same time, then they have a co-occurrence relationship. The co-occurrence matrix is ​​a two-dimensional matrix, and the rows and columns of the matrix represent different words. The element values ​​in the matrix represent the number of co-occurrences of the corresponding two words. For example, if the words "policy" and "implementation" frequently appear together in certain sentences in a news article, the values ​​of the elements corresponding to the "policy" row and "implementation" column in the co-occurrence matrix will increase accordingly. By constructing a co-occurrence matrix, the relationship between words in the text can be presented in a quantitative form.

[0029] Assume that the news text is "AA actively introduces new regulations to vigorously promote economic development. The implementation of the new regulations will effectively stimulate market vitality." Count the number of nouns N n =5 (AA, new regulations, economy, market, vitality), verb number N v =4 (introduce, promote, implement, stimulate), number of adverbs N adv =3 (positive, vigorous, effective).

[0030] Table 1 takes some words as an example to construct the co-occurrence matrix as follows:

[0031] Keywords AA New regulations promote AA 0 1 0 New regulations 1 0 1 promote 0 1 0

[0032] S232 extracts keyword weights based on the contribution matrix. In this embodiment, keyword weights are extracted using the PageRank algorithm. The PageRank algorithm is used to rank web pages. Its core idea is that the importance of a web page depends not only on the number of links pointing to it, but also on the importance of the web pages from which these links come. In keyword weight analysis, the co-occurrence matrix is ​​regarded as a graph structure, the words are the nodes in the graph, the co-occurrence relationship is the edge between the nodes, and the weight of the edge is the number of co-occurrences. The PageRank algorithm assigns an importance score to each word node through iterative calculation, and this score is the keyword weight of the word. The higher the score, the more important the word is in the text. Continuing with the above example, the co-occurrence matrix is ​​used as the adjacency matrix of the graph, and the PageRank algorithm is used for iterative calculation. Assume that after calculation, the keyword weight of "new regulations" is higher, and words such as "AA" and "promote" also have certain weights.

[0033] S233 utilizes the formula: Calculate the color B channel value; wherein B is the color B channel value, ω3 is the verb weight, ω4 is the noun weight, ω5 is the weight of the remaining parts of speech, and m is the total number of words. Through the above steps, the keyword weight information in the news text is converted into the B channel value in the RGB color space, thereby realizing the visual expression of the text information. This embodiment further provides a step for calculating the color B channel, including calculating the color B channel value using the formula: B=256X; wherein B is the color B channel value, X is the proportion of a certain part of speech, and 256 is the total number of B channels. The B value that can also be calculated can be used to determine the color intensity of the blue channel of the news in the visual display.

[0034] S3 aggregates the mean values ​​of the color R channel, the color G channel, and the color B channel of all news to obtain the corresponding news color. Specifically, in this embodiment, the step of aggregating the mean values ​​of the color R channel, the color G channel, and the color B channel of all news within a certain time period, within a certain region, or related to a certain target object into the corresponding news color is to use the formula: Calculate news color; RGB 聚合 is the RGB value of the aggregated news color, ∑R i is the total value of the color R channel of all news, ∑G i is the total value of the color G channel of all news, ∑B i is the total value of the color B channel for all news items, and N is the total number of news items; i = 1, 2, ..., N. This step uses arithmetic averaging to eliminate noise from individual news items and highlight overall trends, such as the proportion of news items of a certain type, the mean sentiment, and the average strength of keyword weights. For example, if the mean R channel for a certain time period or region is between 0 and 12, this indicates that the majority of news in that time period or region is real-time political news. If the mean G channel for a certain time period or region is between 0 and 128, this indicates that the news in that time period or region is mostly negative. The time window in this step can be hourly, daily, or weekly. The hourly level is used to capture real-time fluctuations in breaking news, such as the concentrated release of financial news at noon. The daily level is used to reflect macro-level changes in daily news topics, such as the difference in news types between weekdays and weekends. The weekly level is used to reveal long-term trends, such as whether the mean sentiment value of "environmental protection" news within a certain week is positive. Similarly, the region can be a city, a region within a city, or a country, while the target object can be a company, a celebrity, or an event.

[0035] Table 2 takes the time window as an example. Assuming there are three news items on a certain day, their RGB values ​​are as follows:

[0036] serial number News Type Confidence Positive probability Keyword weight R-value G value B value 1 1(Current Affairs and Politics) 0.8 0.9 0.6 2 230 80 2 5 (Economy) 0.9 0.6 0.7 53 154 90 3 10 (Technology) 0.7 0.7 0.8 113 179 100

[0037] Then, according to the above formula, the final aggregated news color of the day is obtained: RGB(56,188,90).

[0038] In other embodiments, each news type can be aggregated to obtain the news color of the news type, that is,

[0039] Calculate the news color of a certain type of news in a certain period of time or a certain area; similarly, RGB 聚合 is the RGB value of a certain type of news color in a certain period of time or a certain area, R j is the color R channel value of the jth news in a certain type of news in a certain period of time or a certain region, and ∑R j is the total value of the color R channel of n news items of this type in a certain period of time or in a certain region, G j is the color G channel value of the jth news in a certain type of news, and ∑G j is the total value of the color G channel of n news of this type in the period or region, B jis the color G channel value of the jth news in a certain type of news in a certain period of time or in a certain area, ∑B j is the total value of the color B channel of n news of this type in the period or region, and n is the total number of news in the period or region. For example, the method of this embodiment is used to analyze all news data from People's Daily on January 22, 2025, and the three attribute features: news type, sentiment tendency and keyword weight are mapped to R channel, G channel and B channel respectively, and the corresponding news color is generated by aggregation based on the mean value of the three channels: see Figure 7a , for all current affairs and political news from People's Daily Online on January 22, 2025, the mean value is 1 when mapped to the R channel; the mean value is 3 when the sentiment tendency of all real-time political news is mapped to the G channel; the mean value is 91 when the keyword weight of all real-time political news is mapped to the B channel; the three are aggregated to obtain the news color of all current affairs and political news from People's Daily Online on January 22, 2025 as RGB(1391). Figure 7b For all health and medical news from People's Daily Online on January 22, 2025, the average value of mapping them to the R channel is 213; the average value of mapping the sentiment tendency of all real-time political news to the G channel is 212, and the average value of mapping the keyword weights of all real-time political news to the B channel is 68; aggregating the three, the news color of all health and medical news from People's Daily Online on January 22, 2025 is RGB (21321268). For example Figure 8For all current affairs and political news from People's Daily Online on April 20, 2025, the mean value mapped to the R channel is 6; the sentiment of all real-time political news is mapped to the G channel with a mean value of 173; and the keyword weights of all real-time political news are mapped to the B channel with a mean value of 121. Aggregating these three, the news color of all current affairs and political news from People's Daily Online on April 20, 2025, is RGB (6173121). Time-dimensional color aggregation bridges the gap between the micro-features of individual news items and macro-trends over time. Through simple and efficient mean calculation, it transforms massive amounts of news data into interactive color signals, providing an intuitive tool for public opinion monitoring and hotspot tracking. Its core advantage lies in replacing data interpretation with visual perception, significantly improving the efficiency of information transmission. Furthermore, to study news changes and trends over time within a certain time period or region, the news color for each sub-time period within that time period can be periodically obtained. For example, to study news changes from a certain website or multiple websites within a month, the news color for that week can be calculated every other week, and then the news colors for the four weeks can be compared. At the same time, the news color for that month can also be calculated. For another example, to analyze news changes from a certain website or multiple websites within a certain year, the news color for that month can be calculated monthly, and then the news colors for each month can be compared. Of course, the news color for that year will also need to be calculated.

[0040] Example 2: See Figure 1b The present invention also provides another color matching method, which includes the various steps in the above embodiment. The difference is that, after obtaining the news color after the aggregation of N news pieces in step S3, this embodiment further includes the steps of: obtaining the RGB value of a specified color reflecting the theme atmosphere in the current projection scene, and performing secondary aggregation on the specified color and the aggregated news color, and using the aggregated color obtained by the secondary aggregation as the target color to be projected.

[0041]

[0042] Perform secondary polymerization to obtain the target color. 指定 is the R channel value of the specified color, G 指定 is the G channel value of the specified color, B 指定 is the B channel value of the specified color; β is the aggregation factor between the specified color and the aggregated color, and β=0-1.

[0043] Preferably, if the number of all news within a specified time period (for example, this week) or within a specified area (for example, a city) or associated with a specified target object (for example, related to a company) reaches a preset threshold, the value range of β is 0.7 to 0.9 (preferably, β = 0.8) to maintain data dominance; if the number of all news is less than or equal to the preset threshold, the value range of β is 0 to 0.7, and the smaller the number, the smaller the value.

[0044] For some application scenarios, different activities may be held in the same place, such as sculpture exhibitions, art exhibitions, or other exhibitions, and different activities may have different themes. In order to fit the theme, a color will be specified to decorate the venue, for example, accessories of the same color or similar to the specified color, or the specified color will be used as the background color of the entire venue (of course, the specified color does not mean that all objects or backgrounds on the scene are the same color, but refers to the color used for most decorations or backgrounds on the scene). For example, if the current exhibition is on the theme of environmental protection, it is hoped that the background color or ground of the entire venue will be grass green (that is, the specified color), or a sculpture or background curtain will be set up at the venue, and then the corresponding color will be projected onto the sculpture or background curtain through a projector. However, in some scenarios, it is not only necessary to project colors, but also to reflect the overall news situation of the current or local area (or a specified time period or a specified area). Therefore, in this case, not only the color matching problem needs to be considered, but also the degree of fit (or similarity) between the news color obtained after aggregation and the specified color in the scene needs to be considered. Therefore, in this embodiment, the news color obtained by aggregation in step S3 is fused with the specified color in the scene for a second time.

[0045] In actual application, since the mean of each attribute feature is used as the three channel values ​​in step S3, the news associated with the topic may be "diluted" by the mean calculation, thereby affecting the degree of fit or matching between the aggregated color (including the color obtained by the first aggregation and the second aggregation) and the specified color to a certain extent. Therefore, unlike step S2 in embodiment 1, this embodiment, before calculating the R channel value, G channel value, and B channel value of each news, also includes the step of obtaining the correlation between each type of news and the theme atmosphere. Accordingly, the step of calculating the R channel value of each news specifically includes the steps of: if it is highly correlated, using the formula: Calculate the color R channel value; if it is low correlation, use the formula: Calculate the color R channel value. Similarly, the steps for calculating the G channel value of each news item include the following steps: If there is a high correlation, use the formula: Calculate the color G channel value; if it is low correlation, use the formula: ω2×(1-q)] calculates the color G channel value. Similarly, the steps for calculating the B channel value of each news item include the following steps: If it is highly correlated, use the formula: Or B=c1*256X to calculate the color B channel value; if it is low correlation, use the formula: Or B=c2*256X to calculate the color B channel value. Wherein, a1 and a2, b1 and b2, c1 and c2 are pre-set weighting coefficients, and a1>a2, b1>b2, c1>c2. Preferably, a1>1, 0<a2<1; b1>1, 0<b2<1; c1>1, 0<c2<1. That is, by strengthening the color of news that is highly relevant to the topic and weakening the color of news that is less relevant to the topic, the color of the final aggregated news is more closely aligned with or close to the specified color required by the topic; at the same time, the distinction between the aggregated colors is also increased to a certain extent.

[0046] In some embodiments, the correlation between the above news and the theme can be identified by an AI algorithm. Of course, it can also be determined by calculating the number of keywords related to the theme in each news. For example, the theme of an activity is environmental protection. The NLP tool can be used to obtain the number Q of environmental protection-related keywords in each news. If Q is greater than the preset quantity threshold Q0, it is marked as high correlation. If Q is less than or equal to the preset quantity threshold Q0, it is marked as low correlation. Furthermore, if the categories of the Q keywords in a news article appear in ≥3 different categories (environmental problem category, policy category, action category), for example, the three keywords related to environmental protection appearing in the text are: glacier melting (environmental problem), IPCC (subject), compensation (action)); and, among them, the categories appearing in the title / lead are ≥1, and they are core environmental problems or policy keywords, then the news is marked as highly correlated, otherwise it is marked as weakly correlated. Of course, the relevance can be further graded. For example, on the basis of satisfying Q>Q0: the category of the keywords appearing in the text is ≥3, and the category of the keywords appearing in the title / lead is ≥1, and they are core environmental issues or policy keywords, then the news is level I association, that is, strong association; the category of the keywords appearing in the text is ≥2 but less than 3, and the category of the keywords appearing in the title / lead is ≥1, and they are core environmental issues or policy keywords, then the news is level II association, that is, medium association strength; the category of the keywords appearing in the text is <2, then the news is level III association, that is, weak association. Correspondingly, different weighting coefficients can be set for news of each association level. For example: for news of level I association, the weighting coefficients of its three channels are a1, b1, and c1; for news of level II association, the weighting coefficients of its three channels are a3, b3, and c3; for news of level III association, the weighting coefficients of its three channels are a2, b2, and c2; among which, a1>a3>1, 0<a2<1; b1>b3>1, 0<b2<1; c1>c3>1, 0<c2<1.

[0047] In other embodiments, in actual applications, after entering and exiting the secondary aggregation, the deviation between the aggregated color and the specified color may become larger. Therefore, it is necessary to verify the color after the secondary aggregation. Specifically, the method of this embodiment also includes the steps of: calculating the second Euclidean distance between the aggregated color obtained by the secondary aggregation and the specified color; and judging whether the second Euclidean distance is within a preset Euclidean distance threshold range. If so, it means that the color after the secondary aggregation is close to the specified color. Therefore, the color after the secondary aggregation is used as the target color to be projected onto the target object (for example, a sculpture or a curtain) (of course, preferably, if the first Euclidean distance is less than the second Euclidean distance and both are within the Euclidean distance threshold range, the color of the news obtained by the first aggregation is directly used as the target color); otherwise, it means that the color after the secondary aggregation deviates from the specified color, and judging whether the first Euclidean distance between the news color obtained by the first aggregation and the specified color is within a preset Euclidean distance threshold range (for example, 20-40). If so, reducing the above-mentioned aggregation factor β to obtain a new β′=min(β, maximum value in the Euclidean distance threshold range / first Euclidean distance). Specifically, if the current β is greater than 0.6, gradually reduce β (for example, increase it by 0.1 or 0.05 each time). If the first Euclidean distance is not within the Euclidean distance threshold range, that is, the deviation between the color obtained by the first aggregation and the specified color is relatively large, and when the first Euclidean distance is greater than or equal to the preset distance threshold (for example, 100), perform segmented aggregation: set the initial value of β (for example, 0.2), and then use the above-mentioned aggregation formula for secondary aggregation, and then increase the value of β (for example, 0.8), and use the secondary aggregated color and the specified color for tertiary aggregation, so as to avoid losing too many news features in a single aggregation. Of course, calculate the Euclidean distance between the color after segmented aggregation and the specified color again. If it is within the Euclidean distance threshold range, the color obtained by segmented aggregation is used as the target color. Otherwise, readjust the aggregation factor again (for example, increase the initial value of the aggregation factor, and reduce the aggregation factor during the tertiary aggregation) and re-segment aggregation, and repeat this process until the target color is obtained. If both the first and second Euclidean distances are outside the preset Euclidean distance threshold, and the first Euclidean distance is less than the preset distance threshold, the mapping relationship between each news feature and each channel is adjusted. Specifically, the channel value corresponding to each strongly associated news item is recalculated based on the optimized mapping relationship. The optimized R channel calculation formula is: Wherein, α+γ=1, preferably, α=0.6, γ=0.4; the optimized G channel calculation formula is: Wherein, ρ+σ=1, preferably, ρ=0.7, σ=0.3; the optimized B channel calculation formula is: Among them, θ1+θ2+θ3=1, preferably, θ1=0.5, θ2=0.2, and θ3=0.3. Of course, after adjusting the mapping relationship and performing the above steps, the final secondary aggregated color needs to be verified again. If the verification passes (i.e., the second Euclidean distance between the two is within the Euclidean distance threshold), then stop. Otherwise, adjust the weighting coefficients in the above formula again: α, γ, ρ, σ, θ1, θ2, θ3.

[0048] Generally speaking, the weight of news related to the topic is increased, so the color of the news obtained by the first aggregation is not much different from the specified color, or is relatively close to the specified color, that is, the Euclidean distance between the color of the news obtained by the first aggregation and the specified color is within the preset threshold range, but in order to better fit (or to further shorten the Euclidean distance), it is aggregated for the second time. However, in fact, since the color of the first aggregation is also affected by the number of news and is averaged, there may be a certain deviation between the color obtained by the first aggregation and the specified color (for example, the calculated Euclidean distance is not within the above-mentioned preset Euclidean distance threshold range), but after the second aggregation, the Euclidean distance between the obtained aggregated color and the specified color may be within the preset threshold range, or may not be, or may even deviate more. Therefore, in this embodiment, the above-mentioned verification step needs to be performed and adjusted according to the verification result, so that the aggregated color is closer to the specified color and does not lose too much data.

[0049] Although projecting only the final fused color eliminates the need for frequent sculpture changes and allows for projecting different fused news colors based on the actual circumstances of each banquet, projecting only the final fused news color would be too monotonous for artists and connoisseurs, and the information obtained would lack a certain degree of granularity. In view of this, in other embodiments, the designated time period is divided into multiple sub-time periods (for example, if the designated time period is the current week, each sub-time period can be a day within the week), and the mean of the three primary color channels within each sub-time period is generated. These are then aggregated to obtain the news color corresponding to each sub-time period and the news color corresponding to the designated time period. The news colors corresponding to the sub-time periods are then sorted in chronological order (or other specified order), with the news color for the designated time period being ranked first or last before being projected. This not only reflects the current news dynamics (i.e., the designated time period), but also reflects the news dynamics of each sub-time period within the designated time period, while also making the projected colors more colorful.

[0050] Example 3: Figure 2As shown, this embodiment provides a multimodal RGB intelligent color matching system for news text, including: a news data capture module for acquiring news data; an RGB value calculation module for calculating the color R channel based on the news type in the news data; calculating the color G channel based on the news sentiment in the news data; and calculating the color B channel. Specifically, the calculation method for calculating the three channel values ​​of each news item in the RGB value calculation can refer to the calculation method in the above-mentioned embodiment 1 and will not be repeated here; a color aggregation module for aggregating the mean value of the color R channel, the mean value of the color G channel, and the mean value of the color B channel of all specified news items into the news color for that time period. The aggregation method performed by the color aggregation module refers to the aggregation method in the above-mentioned embodiment 1 and will not be repeated here. In other embodiments, the intelligent color matching system also includes: a secondary aggregation module for obtaining the RGB value of the specified color in the current projection scene and performing secondary aggregation with the news color obtained by the color aggregation module to obtain the target color. Specifically, the secondary aggregation method refers to the aggregation method in the above-mentioned embodiment 2 and will not be repeated here. Accordingly, in order to reduce the dilution effect of the mean calculation and thereby enhance the presentation of news associated with the topic in the target color, in other embodiments, the system further includes a correlation acquisition module for acquiring the correlation between each piece of news and the topic. Accordingly, the above-mentioned RGB value calculation module calculates the three channels of each piece of news with different correlation strengths based on the correlation. The specific calculation method can refer to the method in the above-mentioned embodiment 2 and will not be repeated here.

[0051] Example 4: Based on the same inventive concept as above, the present invention also provides another method for mapping the attribute features of news into a single primary color in multimodal RGB intelligent color matching. Specifically, as mentioned above, in order to map the news text into RGB news color, it is necessary to calculate the specific value of the single primary color, thereby mapping the key attribute features of the news into the corresponding single primary color, thereby realizing the mapping of news color. Specifically, see Figure 3, the method of this embodiment includes the steps of: S301 obtaining news data and its attribute features; S302 quantizing each attribute feature, and mapping the quantized value thereof to the channel value of the specified single primary color according to the single primary color calculation specified in advance for any attribute feature; S303 calculating the mean value of any attribute feature of all news mapped to the specified single primary color. In step S301, as described in the above embodiment 1, news data can be captured from the news platform by crawler technology and stored in a MySQL database; the captured fields include id, title, content, time, url, which will not be repeated here. Furthermore, the news data is classified using an SVM or BERT model. Furthermore, sentiment tendency is predicted using one of the Logistic regression model, LSTM or Transformer. As mentioned above, the attribute features of news include: news type, news sentiment tendency, and keyword weight. Therefore, in actual applications, the user can specify a single primary color in RGB for any of the attribute features. In step S302, the step of quantifying the news type specifically includes: removing stop words from each news item in the news data and extracting features; predicting the news type and confidence level of the news item based on the features; and using the formula: Calculate the quantitative value L1 of the news type; where k is the serial number of the news type of the news, n is the number of news types, p is the confidence level, and 256 is the total number of single primary color channels.

[0052] In some embodiments, referring to Example 1, the news can be sorted based on the authority of the source of each news (for example, from mainstream media, or from self-media, and accordingly, the authority of the source from mainstream media is greater than that of self-media), as well as the number of times each news is reprinted (preferably, the sum of the number of times each news of this type is reprinted in all news within a specified time period or a specified area), and coverage. For example, by calculating the importance of each news = 0.4*α+0.3*β+0.3*γ, where if it comes from mainstream media, the source weight α=1.0, if it comes from self-media, α=0.3-0.6; β is the number of times the news is reprinted; γ is the total number of regions covered when the news is reprinted. Then calculate the importance of each news (that is, the sum of the importance of all news of this type in N news) and sort them to obtain the serial number k of each news type.

[0053] In step S302, the step of quantifying the sentiment of the news includes: performing Chinese word segmentation on each news item in the news data and converting it into a vector; predicting the sentiment of the news item based on the vector; and using the formula: Calculate the quantitative value of the news sentiment. Where q is the positive probability, ω1 is the positive weight, ω2 is the negative weight, and 256 is the total number of single-color channels. In step S302, the steps of quantifying the keyword weights include: after segmenting and annotating each news item in the news data, constructing a co-occurrence matrix; extracting keyword weights based on the contribution matrix; using the formula: Calculate the quantitative value of the keyword weight. Where ω3 is the verb weight, ω4 is the noun weight, ω5 is the weight of other parts of speech, and m is the total number of words.

[0054] Correspondingly, if the single primary color specified for the news type is the R primary color, the value of L1 is assigned to the notification value of the R color, that is, R=L1; if the single primary color specified for the news emotional tendency is the R primary color, the value of L2 is assigned to the channel value of the R color, that is, R=L2; if the single primary color specified for the keyword weight is the R primary color, the value of L3 is assigned to the channel value of the R color, that is, R=L3; then the mean of mapping any attribute feature of all news to the specified single primary color is calculated as the mean of L1 corresponding to all news, or the mean of L2 corresponding to all news, or the mean of L3 corresponding to all news; or, if the single primary color specified for the news type is the G primary color, the value of L1 is assigned to the notification value of the R color, that is, G=L1; if the single primary color specified for the news emotional tendency is the G primary color, the value of L2 is assigned to the channel value of the G color, that is, G=L2; if the single primary color specified for the keyword weight is the G primary color, the value of L3 is assigned to the channel value of the G color, that is, G=L2; If the primary color is G, the value of L3 is assigned to the channel value of G, that is, G=L3; then the mean of mapping any attribute feature of all news to the specified single primary color is calculated as the mean of L1 corresponding to all news, or the mean of L2 corresponding to all news, or the mean of L3 corresponding to all news; or, if the single primary color specified for the news type is B, the value of L1 is assigned to the notification value of B, that is, B=L1; if the single primary color specified for the news emotional tendency is B, the value of L2 is assigned to the channel value of B, that is, B=L2; if the single primary color specified for the keyword weight is B, the value of L3 is assigned to the channel value of B, that is, B=L3; then the mean of mapping any attribute feature of all news to the specified single primary color is calculated as the mean of L1 corresponding to all news, or the mean of L2 corresponding to all news, or the mean of L3 corresponding to all news. In some embodiments, the news in step S303 refers to: the desired news within a specified time period, or all news in a specified area, or the desired news related to a specified target object. Specifically, a certain time period includes: a certain day, or a certain week, or a certain month, or a certain year. A certain area includes a certain city, a certain area in a city, a certain province, a certain country, etc. A certain target object includes a certain enterprise, a certain unit, a certain celebrity, a certain event, etc. In this embodiment, each attribute feature of a certain news is quantized to obtain a corresponding quantization value, and then mapped to a specific single primary color channel value, thereby realizing mapping any attribute feature to a specified single primary color, and then taking the mean of the quantization values ​​of the attribute feature of all news as the corresponding single primary color mean, thereby facilitating the comprehensive visualization of all news.

[0055] In order to reduce the diluting effect of the mean on news associated with the topic, in other embodiments, the calculation formulas of the above three channels are improved according to the correlation between the news and the topic. Specifically, referring to the method in the above embodiment 2, different weighting coefficients are set for highly correlated and lowly correlated news, respectively, which will not be repeated here. For example, if the single primary color specified for the news type is the R primary color, then for strongly correlated news: R = a1*L1, and for weakly correlated news: R = a2*L1; if the single primary color specified for the news sentiment tendency is the R primary color, then for strongly correlated news: R = a1*L2, and for weakly correlated news: R = a2*L2; if the single primary color specified for the keyword weight is the R primary color, then for strongly correlated news: R = a1*L3, and for weakly correlated news: R = a2*L3. For another example, if the single primary color specified for news type is G, then for strongly related news: G = b1*L1, and for weakly related news: R = b2*L1; if the single primary color specified for news sentiment is G, then for strongly related news: G = b1*L2, and for weakly related news: R = b2*L2; if the single primary color specified for keyword weight is G, then for strongly related news: G = b1*L3, and for weakly related news: R = b2*L3; For example, if the single primary color specified for the news type is B primary color, then for strongly associated news: G = c1*L1, and for weakly associated news: R = c2*L1; if the single primary color specified for the news emotional tendency is B primary color, then for strongly associated news: G = c1*L2, and for weakly associated news: R = c2*L2; if the single primary color specified for the keyword weight is B primary color, then for strongly associated news: G = c1*L3, and for weakly associated news: R = c2*L3.

[0056] Example 5: The present invention also provides another method for mapping the attribute features of news into a single primary color in multimodal RGB intelligent color matching, which includes the steps in the above-mentioned Example 4. The difference is that in this embodiment, the user pre-specifies the corresponding single primary color for each attribute feature. Figure 4Specifically, it includes the following steps: S401 identifies the type of the current user, if the current user is a new user, executes step S402, if the current user is an old user, executes step S403; S402 displays six preset mapping relationship templates to the current user; wherein, the most frequently used one is the benchmark reference template, executes step S404; S403 calculates the preference index Ti=wi*(Ni / M) of each template based on the frequency of each template used by the old user, and displays at least one template with a preference index greater than a preset threshold and the benchmark reference template to the old user, executes step S404; the benchmark reference template is the most frequently used template among the six preset mapping relationship templates; wherein Ti is the first The preference index of i templates, Ni is the number of times the i-th template is used by the old user, M is the total number of times all templates are used by the old user; wi is the weight of each template; S404 When the user arbitrarily specifies a template, and the specified template is not the benchmark reference template, the corresponding example news colors are generated based on the benchmark reference template and the template selected by the user, as well as all news in the preset historical time period (or in a certain area, or related to a certain target object); wherein the preset historical time period is the most recent week or the most recent month; S405 When the user selects an example news color, its corresponding template is used as the template specified by the user to obtain the mapping relationship between the news attribute features and the three primary colors, and execute step S302. As mentioned above, the attribute features of news include: news type, news emotional tendency, and keyword weight, and RGB also includes three primary colors. Therefore, in actual applications, the mapping relationship between attribute features and primary colors can be determined according to the type of user. If it is a new user (i.e., a primary user), six preset mapping relationship templates can be displayed to the user, such as Figure 6a-6fAs shown: First template: News type is mapped to color R, sentiment is mapped to color G, and keyword weight is mapped to color B; Second template: News type is mapped to color R, sentiment is mapped to color B, and keyword weight is mapped to color G; Third template: News type is mapped to color G, sentiment is mapped to color R, and keyword weight is mapped to color B; Fourth template: News type is mapped to color G, sentiment is mapped to color B, and keyword weight is mapped to color R; Fifth template: News type is mapped to color B, sentiment is mapped to color G, and keyword weight is mapped to color R; Sixth template: News type is mapped to color B, sentiment is mapped to color R, and keyword weight is mapped to color G. By presenting six preset templates to users, they can choose according to their preferences to construct different news colors. Of course, the most frequently used template can also be marked as a reference template for user reference. Preferably, the most frequently used template here refers to the template that has been used the most times by other users within a preset historical time period based on big data analysis. Because users cannot predict the final news color effects mapped from their selected templates (partly because most users lack knowledge of the three primary colors and therefore cannot predict them, and partly because the attributes of news text vary over different time periods), corresponding sample news colors can be generated based on all news within a preset historical time period. In other words, by presenting the news color effects to users through a preview, users are given the opportunity to select from a specified template and a reference template, improving the user experience.

[0057] Furthermore, the time interval between the preset historical time period and the time period to be analyzed by the user is less than the preset time interval, or is a part of the time period. This is to avoid the situation where the previewed news color is too different from the news ultimately obtained based on the user-specified template (i.e., the attribute characteristics change too much), which may not meet the user's expectations and thus reduce the user experience. For example, if the user wants to analyze all news in March 2025, the preset historical time period is February 2025, or early March 2025 (this requires less computation and generates the previewed news color faster). Of course, in other embodiments, the preset historical time period may also be March 2025.

[0058] In other embodiments, if the user is an old user, especially an old user who has used every template, he or she usually has his or her own preferred color matching template. Therefore, if the old user is identified, one method is to recommend the color matching template used most recently, or the one with the highest frequency of use, by default. However, this method is only suitable for senior customers (i.e., level I users) who have used the system for a very long time, and the user has used a certain template for a long time, that is, the frequency of use is much greater than the frequency of use of other templates. However, in actual applications, there are some users (i.e., level II users) whose use time is not too long, or who like to try various new things, for example, they like to try different mapping relationship templates. Therefore, this method is not suitable for the above method. Therefore, in this embodiment, the user's preference index for each template is calculated, and at least one template is recommended to the user based on the preference index. Specifically, the preference index is based on the percentage of the number of times the user uses each template. In other embodiments, preference indexes may not be used, and instead the user may be presented with historical news colors from the historical data that the user has already analyzed. Thus, when the user specifies a historical news color, the corresponding mapping template is retrieved as the user-specified template, and steps S405 / S404 are executed. Furthermore, if at least one template with a preference index greater than a preset threshold in step S403 does not include a baseline reference template, and the template ultimately specified by the user is a baseline reference template, the preference index of each template is updated based on the provided result. For example, the preference index of each template is recalculated based on the formula Ti = Ni / M.

[0059] Example 6: Based on the above-mentioned method of mapping news types into single primary colors in multimodal RGB intelligent color matching, the present invention also provides a device for mapping news types into single primary colors in multimodal RGB intelligent color matching, which specifically includes: a first data acquisition module, used to acquire news data and its attribute characteristics; the attribute characteristics include: news type, news sentiment tendency, keyword weight; a single primary color calculation module, used to calculate the channel value of mapping the news data into a specified single primary color according to any attribute characteristic of the news data and the single primary color pre-specified for it; and calculate the mean value of mapping the attribute characteristics of all news into the specified single primary color; wherein all news refers to: all news within a specified time period, or all news in a specified area, or the required news related to a specified target object; specifically, a certain time period includes: a certain day, or a certain week, or a certain month, or a certain year; a certain area includes a certain city, a certain area in a city, a certain province, a certain country, etc.; a certain target object includes a certain enterprise, a certain unit, a certain star, a certain event, etc.

[0060] The single-primary color calculation module specifically includes: an R channel calculation unit, which is used to remove stop words from each news item in the news data and extract features when the single-primary color specified for the news type is R color; then predict the news type of the news item based on the features; and use the formula: Calculate the color R channel value; where R is the color R channel value, K is the K-th type of news, n is the number of news types, p is the confidence level, and 256 is the total number of R channels; or, when the single primary color specified for the news sentiment is R color, perform Chinese word segmentation on each news item in the news data and convert it into a vector; then predict the sentiment of the news item based on the vector; and use the formula: Calculate the color R channel value; where R is the color R channel value, q is the positive probability, ω1 is the positive weight, ω2 is the negative weight, and 256 is the total number of R channels; or, when the single primary color specified for the keyword weight is R color, segment each news item in the news data and mark the part of speech, then construct a co-occurrence matrix; then extract the keyword weight based on the contribution matrix; and use the formula: Calculate the color R channel value; where R is the color R channel value, ω3 is the verb weight, ω4 is the noun weight, ω5 is the weight of the remaining parts of speech, and m is the total number of words.

[0061] In some embodiments, the device further includes: a user type identification module for identifying the type of the current user; the types include new users and old users; an interaction module for displaying six preset mapping relationship templates to the current user when it is identified that the current user is a new user, and when the current user arbitrarily specifies a template, and the specified template is not the baseline reference template, generating corresponding example news colors based on the baseline reference template and the template selected by the user, as well as all news in a preset historical time period; wherein the baseline reference template is the most frequently used one of the six templates; the preset historical time period is the most recent week or the most recent month; the module The template designation module is used to, when the current user selects an example news color, use its corresponding template as the user-designated template to obtain the mapping relationship between the three attribute features and the three primary colors; the template recommendation module is used to, when it is identified that the current user is an old user, calculate the preference index Ti = Ni / M of each template based on the frequency of each template used by the old user, and trigger the interaction module to display at least one template with a preference index greater than a preset threshold and a benchmark reference template to the old user; wherein Ti is the preference index of the i-th template, Ni is the number of times the i-th template is used by the old user, and M is the total number of times all templates are used by the old users.

[0062] In other embodiments, the single primary color calculation module is further used to obtain the correlation between each news item and the theme atmosphere, and when the correlation is high, the calculation formula of the R channel value, the G channel value, and the B channel value is optimized using the weighted coefficient: If the correlation is low, use the weighted coefficients to optimize the calculation formulas for the R, G, and B channel values: Or B=c1*256X; wherein a1 and a2, b1 and b2, c1 and c2 are preset weighting coefficients, and a1>a2, b1>b2, c1>c2.

[0063] Example 7: The present invention also provides an RGB intelligent color matching method based on news text, specifically, see Figure 5 The method includes: S501 obtaining news data and attribute features thereof; the attribute features include: news type, or news sentiment, or keyword weight; S502 obtaining a mapping relationship template between news attribute features and three primary colors preset by a user; the mapping relationship template includes: news type is mapped to a first primary color, sentiment is mapped to a second primary color, and keyword weight is mapped to a third primary color; wherein the first primary color, the second primary color, and the third primary color are respectively one of R color, G color, and B color; S503 calculating the color channel values ​​of each of the three primary color channels based on the mapping relationship template and the news attribute features; if the first primary color is R color, the second primary color is G color, and the third primary color is B color, the step of calculating the color channel values ​​of each of the three primary color channels specifically includes: calculating the color R channel according to the news type in the news data; calculating the color G channel according to the news sentiment in the news data; and calculating the color B channel according to the keyword weight; S504 aggregating the average value of the color R channel, the average value of the color G channel, and the average value of the color B channel of all news into a corresponding aggregated news color. Wherein, step S504 is to use the formula: Calculate the aggregated news color; where RGB 聚合 is the RGB value of the aggregated news color, R i is the R channel value of the i-th news in N news, ∑R i is the total value of the R channel of the N news, G i is the G channel value of the i-th news in N news, B i is the B channel value of the i-th news among N news, N is the total number of news, i=1,2···N.

[0064] In some embodiments, in step S502, when the user is a new user, the step of obtaining the mapping relationship between the news attribute features pre-set by the user and the three primary colors specifically includes: S5021a displaying six preset mapping relationship templates; among which, the most frequently used one is the baseline reference template; S5022a when the user arbitrarily specifies a template, and the specified template is not the baseline reference template, respectively generating corresponding example news colors based on the baseline reference template and the template selected by the user, as well as all news within a preset historical time period; wherein, the preset historical time period is the most recent week or the most recent month; S5023a when the user selects an example news color, its corresponding template is used as the template specified by the user to obtain the mapping relationship between the news attribute features and the three primary colors.

[0065] In some embodiments, in step S5021a, the six preset mapping relationship templates are: the first template: news type is mapped to R color, emotional tendency is mapped to G color, and keyword weight is mapped to B color; the second template: news type is mapped to R color, emotional tendency is mapped to B color, and keyword weight is mapped to G color; the third template: news type is mapped to G color, emotional tendency is mapped to R color, and keyword weight is mapped to B color; the fourth template: news type is mapped to G color, emotional tendency is mapped to B color, and keyword weight is mapped to R color; the fifth template: news type is mapped to B color, emotional tendency is mapped to G color, and keyword weight is mapped to R color; the sixth template: news type is mapped to B color, emotional tendency is mapped to R color, and keyword weight is mapped to G color.

[0066] By presenting six preset templates to users, they can choose according to their preferences to create different news colors. Furthermore, the most frequently used template can be marked as a baseline reference template for user reference. Preferably, the most frequently used template here refers to the template that, based on big data analysis, has been used the most times by other users within a preset historical time period among the six templates.

[0067] Because users cannot predict the final news color effects mapped from their selected templates (partly because most users lack knowledge of the three primary colors and therefore cannot predict them, and partly because the attributes of news text vary over different time periods), corresponding sample news colors can be generated based on all news within a preset historical time period. In other words, by presenting the news color effects to users through a preview, users are given the opportunity to select from a specified template and a reference template, improving the user experience.

[0068] Furthermore, the time interval between the preset historical time period and the time period to be analyzed by the user is less than the preset time interval, or is a part of the time period. This is to avoid the situation where the previewed news color is too different from the news ultimately obtained based on the user-specified template (i.e., the attribute characteristics change too much), which may not meet the user's expectations and thus reduce the user experience. For example, if the user wants to analyze all news in March 2025, the preset historical time period is February 2025, or early March 2025 (this requires less computation and generates the previewed news color faster). Of course, in other embodiments, the preset historical time period may also be March 2025.

[0069] In some embodiments, in step S502, when the user is an old user, the step of obtaining the mapping relationship between news attribute features preset by the user and the three primary colors specifically includes: S5021b calculating a preference index Ti = Ni / M for each template based on the frequency of use of each template by the old user, where Ti is the preference index of the i-th template, Ni is the number of times the i-th template is used by the old user, and M is the total number of times all templates are used by the old user; S5022b displaying at least one template having a preference index greater than a preset threshold and a baseline reference template to the old user; the baseline reference template is the template with the highest frequency of use among the six preset mapping relationship templates; S5023b when the user arbitrarily specifies a template and the specified template is not the baseline reference template, generating corresponding example news colors based on the baseline reference template and the template selected by the user, as well as all news within a preset historical time period; wherein the preset historical time period is the most recent week or the most recent month; S5024b when the user selects an example news color, using its corresponding template as the template specified by the user to obtain a mapping relationship between news attribute features and the three primary colors.

[0070] Correspondingly, if in step S503, if the first primary color is R color, the second primary color is G color, and the third primary color is B color, the color R channel is calculated according to the news type in the news data; the color G channel is calculated according to the news sentiment tendency in the news data; and the color B channel is calculated according to the keyword weight; then: the step of calculating the color R channel according to the news type in the news data specifically includes: removing stop words from each news in the news data and extracting features; predicting the news type to which the news belongs and its confidence p based on the features; using the formula: Calculate the color R channel value; where R is the color R channel value, k is the serial number of the news type of the news in the news type in n, n is the number of news types, p is the confidence level, and 256 is the total number of R channels;

[0071] The steps for calculating the color G channel based on the news sentiment in the news data specifically include: performing Chinese word segmentation on each news item in the news data and converting it into a vector; predicting the sentiment of the news item based on the vector; and using the formula: Calculate the color G channel value; where G is the color G channel value, q is the positive probability, ω1 is the positive weight, ω2 is the negative weight, and 256 is the total number of G channels;

[0072] The steps of calculating the color B channel according to the keyword weight specifically include: after segmenting each news in the news data and marking the part of speech, constructing a co-occurrence matrix; extracting the keyword weight based on the contribution matrix; using the formula: Calculate the color B channel value; where B is the color B channel value, ω3 is the verb weight, ω4 is the noun weight, ω5 is the weight of the remaining parts of speech, and m is the total number of words.

[0073] With reference to Example 1, the news can be sorted based on the authority of the source of each news (for example, from mainstream media, or from self-media, and accordingly, the authority of news from mainstream media is greater than that from self-media), the number of times each news is reprinted (preferably, the sum of the number of times each news of this type is reprinted in all news within a specified time period or a specified area), and the coverage. For example, by calculating the importance of each news = 0.4*α+0.3*β+0.3*γ, where if it comes from mainstream media, the source weight α=1.0, if it comes from self-media, α=0.3-0.6; β is the number of times the news is reprinted; γ is the total number of regions covered when the news is reprinted. Then calculate the importance of each news (that is, the sum of the importance of all news of this type in N news), and sort them to obtain the serial number k of each news type. It can be seen that when the first primary color is R, the second primary color is G, and the third primary color is B, the specific steps of calculating the mean of the three primary colors for all news in a certain time period and finally obtaining the news color can be referred to the above embodiment 1 and will not be repeated here.

[0074] Accordingly, if in step S503, if the first primary color is R color, the second primary color is B color, and the third primary color is G color, the color R channel is calculated according to the news type in the news data; the color B channel is calculated according to the news sentiment tendency in the news data; and the color G channel is calculated according to the keyword weight; then: the step of calculating the color R channel according to the news type in the news data specifically includes: removing stop words from each news in the news data and extracting features; predicting the news type and confidence level of the news according to the features; and using the formula: Calculate the color R channel value; where R is the color R channel value, k is the sequence number of the news type of the news in the news type n, n is the number of news types, p is the confidence level, and 256 is the total number of R channels; the steps of calculating the color B channel based on the news sentiment in the news data specifically include: performing Chinese word segmentation on each news in the news data and converting it into a vector; predicting the sentiment of the news based on the vector; using the formula: Calculate the color B channel value; where B is the color B channel value, q is the positive probability, ω1 is the positive weight, ω2 is the negative weight, and 256 is the total number of B channels; the steps of calculating the color G channel according to the keyword weight specifically include: after segmenting each news in the news data and marking the part of speech, constructing a co-occurrence matrix; extracting the keyword weight based on the contribution matrix; using the formula: Calculate the color G channel value; where G is the color G channel value, ω3 is the verb weight, ω4 is the noun weight, ω5 is the weight of the remaining parts of speech, and m is the total number of words.

[0075] Correspondingly, if in step S503, if the first primary color is G color, the second primary color is B color, and the third primary color is R color, the color G channel is calculated according to the news type in the news data; the color B channel is calculated according to the news sentiment tendency in the news data; and the color R channel is calculated according to the keyword weight; then: the step of calculating the color G channel according to the news type in the news data specifically includes: removing stop words from each news in the news data and extracting features; predicting the news type of the news according to the features; and using the formula: Calculate the color G channel value; where G is the color G channel value, k is the sequence number of the news type of the news in the news type n, n is the number of news types, p is the confidence level, and 256 is the total number of G channels; the steps of calculating the color B channel based on the news sentiment in the news data specifically include: performing Chinese word segmentation on each news in the news data and converting it into a vector; predicting the sentiment of the news based on the vector; using the formula: Calculate the color B channel value; where B is the color B channel value, q is the positive probability, ω1 is the positive weight, ω2 is the negative weight, and 256 is the total number of B channels; the steps of calculating the color R channel according to the keyword weight specifically include: after segmenting each news in the news data and marking the part of speech, constructing a co-occurrence matrix; extracting the keyword weight based on the contribution matrix; using the formula: Calculate the color R channel value; where R is the color R channel value, ω3 is the verb weight, ω4 is the noun weight, ω5 is the weight of the remaining parts of speech, and m is the total number of words.

[0076] Accordingly, if in step S503, if the first primary color is G color, the second primary color is R color, and the third primary color is B color, the color G channel is calculated according to the news type in the news data; the color R channel is calculated according to the news sentiment tendency in the news data; and the color B channel is calculated according to the keyword weight; then: the step of calculating the color G channel according to the news type in the news data specifically includes: removing stop words from each news in the news data and extracting features; predicting the news type of the news according to the features; and using the formula: Calculate the color G channel value; where G is the color G channel value, k is the sequence number of the news type of the news in the news type n, n is the number of news types, p is the confidence level, and 256 is the total number of G channels; the steps of calculating the color R channel based on the news sentiment in the news data specifically include: performing Chinese word segmentation on each news in the news data and converting it into a vector; predicting the sentiment of the news based on the vector; using the formula: The steps of calculating the color R channel value and calculating the color B channel according to the keyword weight include: constructing a co-occurrence matrix after segmenting and marking the part of speech of each news in the news data; extracting the keyword weight based on the contribution matrix; and using the formula: Calculate the color B channel value; where B is the color R channel value, ω3 is the verb weight, ω4 is the noun weight, ω5 is the weight of the remaining parts of speech, and m is the total number of words.

[0077] Correspondingly, if in step S503, if the first primary color is B color, the second primary color is R color, and the third primary color is G color, the color B channel is calculated according to the news type in the news data; the color R channel is calculated according to the news sentiment tendency in the news data; and the color B channel is calculated according to the keyword weight, then: the step of calculating the color B channel according to the news type in the news data specifically includes: removing stop words from each news in the news data and extracting features; predicting the news type of the news according to the features; and using the formula: Calculate the color B channel value; where B is the color B channel value, k is the sequence number of the news type of the news in the news type n, n is the number of news types, p is the confidence level, and 256 is the total number of B channels; the steps of calculating the color R channel based on the news sentiment in the news data specifically include: performing Chinese word segmentation on each news in the news data and converting it into a vector; predicting the sentiment of the news based on the vector; using the formula: Calculate the color R channel value; where R is the color R channel value, q is the positive probability, ω1 is the positive weight, ω2 is the negative weight, and 256 is the total number of R channels; the steps of calculating the color G channel according to the keyword weight specifically include: after segmenting each news in the news data and marking the part of speech, constructing a co-occurrence matrix; extracting the keyword weight based on the contribution matrix; using the formula: Calculate the color B channel value; where G is the color G channel value, ω3 is the verb weight, ω4 is the noun weight, ω5 is the weight of the remaining parts of speech, and m is the total number of words.

[0078] Correspondingly, if in step S503, if the first primary color is B color, the second primary color is G color, and the third primary color is R color, the color B channel is calculated according to the news type in the news data; the color G channel is calculated according to the news sentiment tendency in the news data; and the color R channel is calculated according to the keyword weight, then: the step of calculating the color B channel according to the news type in the news data specifically includes: removing stop words from each news in the news data and extracting features; predicting the news type of the news according to the features; and using the formula: Calculate the color B channel value; where B is the color B channel value, k is the sequence number of the news type of the news in the news type n, n is the number of news types, p is the confidence level, and 256 is the total number of G channels; the steps of calculating the color G channel based on the news sentiment in the news data specifically include: performing Chinese word segmentation on each news in the news data and converting it into a vector; predicting the sentiment of the news based on the vector; using the formula: Calculate the color G channel value; where G is the color G channel value, q is the positive probability, ω1 is the positive weight, ω2 is the negative weight, and 256 is the total number of G channels; the steps of calculating the color R channel according to the keyword weight specifically include: after segmenting each news in the news data and marking the part of speech, constructing a co-occurrence matrix; extracting the keyword weight based on the contribution matrix; using the formula: Calculate the color R channel value; where R is the color R channel value, ω3 is the verb weight, ω4 is the noun weight, ω5 is the weight of the remaining parts of speech, and m is the total number of words. Of course, the news color obtained by the color matching method of this embodiment can also be applied to the projection scene in the second embodiment, that is, the aggregated news color is secondary aggregated with the specified color. The method of secondary aggregation can refer to the above-mentioned second embodiment and will not be repeated here. Furthermore, since it is necessary to fit the projection scene, when mapping any feature of a piece of news to any specified single primary color channel, the method in the second embodiment can be referred to. Based on the correlation between the news and the topic, the corresponding weighting coefficients are set for the mapping formulas of the above three channels.

[0079] Embodiment 8: Based on the color matching method of the above-mentioned embodiment 7, the present invention also provides a multimodal RGB intelligent color matching system for news texts, which includes: a first data acquisition module, used to obtain news data and its attribute characteristics; the attribute characteristics include: news type, or news emotional tendency, or keyword weight; a second data acquisition module, used to obtain a mapping relationship template between news attribute characteristics and three primary colors pre-set by the user; the mapping relationship template includes: news type is mapped to the first primary color, emotional tendency is mapped to the second primary color, and keyword weight is mapped to the third primary color; wherein, the first primary color, the second primary color and the third primary color are respectively one of R color, G color and B color; an RGB value calculation module, used to calculate the color channel value of each of the three primary color channels based on the mapping relationship template and the news attribute characteristics; a color aggregation module, used to aggregate the mean value of the color R channel, the mean value of the color G channel, and the mean value of the color B channel of all news in a certain time period into the news color of the time period.

[0080] The RGB value calculation module specifically includes: an R channel calculation unit, which is used to calculate and remove stop words from each news item in the news data when the first primary color is R color, extract features, and predict the news type of the news item based on the features, and then use the formula: Calculate the color R channel value; where R is the color R channel value, K is the K-th type of news, n is the number of news types, p is the confidence level, and 256 is the total number of R channels; the G channel calculation unit is used to perform Chinese word segmentation processing on each news item in the news data when the second primary color is R color, convert it into a vector, and predict the emotional tendency of the news item based on the vector; then use the formula: Calculate the color G channel value; where G is the color G channel value, q is the positive probability, ω1 is the positive weight, ω2 is the negative weight, and 256 is the total number of G channels; the B channel calculation unit is used to, when the third primary color is color B, segment and mark each news item in the news data, and then construct a co-occurrence matrix; extract keyword weights based on the contribution matrix, and use the formula: Calculate the color B channel value; where B is the color B channel value, ω3 is the verb weight, ω4 is the noun weight, ω5 is the weight of the remaining parts of speech, and m is the total number of words.

[0081] In some embodiments, the system further includes: a user type identification module for identifying the type of the current user; the types include new users and old users; an interaction module for displaying six preset mapping relationship templates to the current user when the current user is identified as a new user, and when the current user arbitrarily specifies a template, and the specified template is not the baseline reference template, generating corresponding example news colors based on the baseline reference template and the template selected by the user, as well as all news in a preset historical time period; wherein the baseline reference template is the most frequently used one of the six templates; the preset historical time period is the most recent week or the most recent month; the template The template designation module is used to, when the current user selects an example news color, use its corresponding template as the user-designated template to obtain the mapping relationship between news attribute features and the three primary colors; the template recommendation module is used to, when it is identified that the current user is an old user, calculate the preference index Ti = Ni / M of each template based on the frequency of each template used by the old user, and trigger the interaction module to display at least one template with a preference index greater than a preset threshold and a benchmark reference template to the old user; wherein Ti is the preference index of the i-th template, Ni is the number of times the i-th template is used by the old user, and M is the total number of times all templates are used by the old users.

[0082] The embodiments of the present invention are described above in conjunction with the accompanying drawings, but the present invention is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the guidance of the present invention, ordinary technicians in this field can also make many forms without departing from the scope of protection of the present invention and the claims, all of which are protected by the present invention.

Claims

1. A multimodal RGB intelligent color matching method for news text, characterized by include: Get news data; Calculate the corresponding color R channel of each news item according to its news type in the news data; Calculate the corresponding color G channel based on the news sentiment of each news item in the news data; calculate the corresponding color B channel of each news item based on the keyword weight of each news item; Aggregate the mean of the color R channel, the mean of the color G channel, and the mean of the color B channel of all news to get the corresponding news color: Get the RGB value of the specified color that reflects the theme atmosphere in the current projection scene, and perform secondary aggregation on the specified color and the aggregated news color to obtain the target color for projection: Where β is the aggregation factor between the specified color and the aggregated color, and β = 0-1; R i is the color R channel value of the i-th news, and ∑R i The total value of the R channel of all news colors, G i is the color G channel value of the i-th news, and ∑G i is the total value of the G channel of all news, B i is the color G channel value of the i-th news, ∑B i is the total value of the color B channel of all news, and N is the total number of news.

2. A multimodal RGB intelligent color matching method for news text according to claim 1, characterized in that The steps to obtain news data include: News data is captured from news platforms using crawler technology and stored in a MySQL database; the captured fields include id, title, content, time, and url.

3. A multimodal RGB intelligent color matching method for news text according to claim 1, characterized in that The steps for calculating the color R channel according to the news type in the news data include: Sort n news types based on the pre-set importance of each news type, and obtain a serial number corresponding to each news type; wherein, the greater the importance, the larger the serial number corresponding to the news type; Remove stop words from each news item in the news data and extract features; Predict the news type and confidence level p of the news based on the features; Using the formula: Calculate the color R channel value; where k is the serial number of the type of news item; and 256 is the total number of R channels.

4. A multimodal RGB intelligent color matching method for news text according to claim 1, characterized in that The steps for calculating the color G channel based on the news sentiment in the news data include: Perform Chinese word segmentation on each news item in the news data and convert it into a vector; Predict the sentiment of the news based on the vector; Using the formula: Calculate the color G channel value; where q is the positive probability that the sentiment of the news is positive, ω1 is the positive weight, ω2 is the negative weight, and 256 is the total number of G channels.

5. A multimodal RGB intelligent color matching method for news text according to claim 1, characterized in that The steps to calculate the color B channel include: After segmenting and marking the part of speech for each news item in the news data, a co-occurrence matrix is ​​constructed; Extracting keyword weights based on the contribution matrix; Using the formula: Calculate the color B channel value; where B is the color B channel value, ω3 is the verb weight, ω4 is the noun weight, ω5 is the weight of the remaining parts of speech, and m is the total number of words.

6. A multimodal RGB intelligent color matching method for news text according to claim 1, characterized in that The steps for calculating the color B channel include using the formula: B = 256X; Calculate the color B channel value; where B is the color B channel value, X is the proportion of a certain part of speech, and 256 is the total number of B channels.

7. A multimodal RGB intelligent color matching method for news text according to any one of claims 1 to 6, characterized in that: Before calculating the color R channel, color G channel, and color B channel values ​​corresponding to each piece of news, the steps further include: obtaining the correlation between each piece of news and the theme atmosphere; If it is highly correlated, use the formula: Calculate the corresponding color R channel value; use the formula: Calculate the corresponding color G channel value; use the formula: Calculate the corresponding color B channel value; If the correlation is low, use the formula: Calculate the corresponding color R channel value; use the formula: Calculate the corresponding color G channel value; use the formula: Or B=c1*256X calculates the corresponding color B channel value; Wherein, a1 and a2, b1 and b2, c1 and c2 are preset weighting coefficients, and a1>a2, b1>b2, c1>c2.

8. A multimodal RGB intelligent color matching method for news text according to claim 1, characterized in that Also includes the steps: Calculate the second Euclidean distance between the color obtained by secondary aggregation and the specified color; Determine whether the second Euclidean distance is within a preset Euclidean distance threshold range. If so, use the color after the second aggregation as the target color; otherwise, calculate the first Euclidean distance between the news color obtained by the first aggregation and the specified color, and determine whether the first Euclidean distance between the news color obtained by the first aggregation and the specified color is within a preset Euclidean distance threshold range. If not, determine whether the first Euclidean distance is greater than or equal to the preset distance threshold. If so, perform segmented aggregation.

9. A multimodal RGB intelligent color matching system for news texts, characterized by include: News data crawling module, used to obtain news data; The RGB value calculation module is used to calculate the color R channel corresponding to each news item in the news data according to its news type; calculate the color G channel corresponding to each news item in the news data according to its news sentiment tendency; and calculate the color B channel corresponding to each news item based on its keyword weight. The color aggregation module is used to aggregate the mean values ​​of the color R channel, the color G channel, and the color B channel of all news to obtain the corresponding news color; The secondary aggregation module is used to obtain the RGB value of the specified color reflecting the theme atmosphere in the current projection scene, and perform secondary aggregation on the specified color and the aggregated news color to obtain the target color for projection.

10. A multimodal RGB intelligent color matching system for news text according to claim 9, characterized in that The RGB value calculation module includes: The R channel calculation unit is used to sort the n news types based on the preset importance of each news type and obtain the serial number corresponding to each news type; then, each news item in the news data is processed to remove stop words, extract features, and predict the news type of the news item based on the features, and then use the formula: Calculate the color R channel value; where R is the color R channel value, k is the serial number of the type to which the news belongs, n is the number of news types, p is the confidence level, and 256 is the total number of R channels; The G channel calculation unit is used to perform Chinese word segmentation on each news item in the news data, convert it into a vector, and predict the sentiment tendency of the news item based on the vector; then use the formula: Calculate the color G channel value; where G is the color G channel value, q is the positive probability, ω1 is the positive weight, ω2 is the negative weight, and 256 is the total number of G channels; Channel B calculation unit is used to segment and mark the part of speech of each news in the news data, and then construct a co-occurrence matrix; Extract keyword weights based on the contribution matrix and use the formula: Calculate the color B channel value; where B is the color B channel value, ω3 is the verb weight, ω4 is the noun weight, ω5 is the weight of the remaining parts of speech, and m is the total number of words.

Images