Scientific and technological information collection method and system based on multi-modal data acquisition

Abstract

The invention discloses a scientific and technological information collection method and system based on multi-modal data acquisition, and relates to the technical field of information data processing, and the main scheme is as follows: collecting texts, images, audios, videos and information basic data, calculating an information timeliness index according to the information basic data, comparing the information timeliness index with a threshold value, screening information data in timeliness, and obtaining a scientific and technological information database; a text index, an image index, an audio index and a video index are calculated according to text data, image data, audio data and video data and compared with respective threshold values to screen and collect standard data, and the system comprises a corresponding data collection module, an index calculation module, a screening module and the like. Therefore, the method has important application value in the field of science and technology intelligence.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligence data processing, and in particular to a method and system for collecting scientific and technological intelligence based on multimodal data acquisition. Background Art

[0002] In the field of scientific and technological intelligence, with the rapid development of information technology, intelligence data has shown a significant trend of explosive growth and multimodality. Multimodal data covers rich text, image, audio and video information. How to efficiently and accurately collect valuable scientific and technological intelligence from massive and complex multimodal data has become a key issue that needs to be solved urgently.

[0003] Existing methods for collecting scientific and technological intelligence often simply search for keywords in intelligence data and collect intelligence data with the same keywords in a unified manner, lacking quality control over the intelligence data.

[0004] Therefore, existing technologies have obvious defects. For example, it is difficult to accurately control the timeliness of intelligence, and it is impossible to timely screen out the latest and most valuable intelligence based on the dynamic changes of data. At the same time, there is a lack of a comprehensive and scientific comprehensive evaluation system, and it is impossible to accurately judge the quality and reliability of intelligence from the overall perspective of multimodal data. As a result, the effectiveness of the collected scientific and technological intelligence in practical applications is greatly reduced, and it is difficult to meet the growing demand for high-quality scientific and technological intelligence in scientific research innovation, corporate decision-making, etc. Summary of the invention

[0005] 1. Technical issues to be resolved In view of the shortcomings of the prior art, the present invention provides a method and system for collecting scientific and technological intelligence based on multimodal data acquisition, which solves the problem of simple and inaccurate intelligence timeliness assessment in the prior art by collecting basic intelligence data, calculating the intelligence timeliness index based on the data, and screening and classifying the multimodal data based on the comparison result with the intelligence timeliness index threshold. The method and system also solve the problem of the prior art in which the intelligence timeliness assessment is simple and inaccurate by separately calculating the indexes corresponding to text, image, audio and video data, and screening the standard data by comparing them with their respective thresholds.

[0006] (II) Technical solution To achieve the above objectives, the present invention is implemented through the following technical solutions: a method and system for collecting scientific and technological intelligence based on multimodal data acquisition, comprising: Collect basic intelligence data, which includes some basic intelligence; According to the basic intelligence data, calculate the intelligence timeliness index TI of several basic intelligences; set the intelligence timeliness index threshold F, and according to the comparison result of the intelligence timeliness index TI and the intelligence timeliness index threshold F, screen and collect the basic intelligence within the timeliness, and classify the screened basic intelligence into text data, image data, audio data and video data; The text index dex is calculated based on the text data, and the image index ma and the average shooting time are calculated based on the image data. , calculate the audio index tex according to the audio data, and calculate the video index vid according to the video data; A basic intelligence threshold set is set, and the text index dex, image index ma, audio index tex, and video index vid are compared with the corresponding thresholds in the basic intelligence threshold set to screen basic intelligence that meets the standards.

[0007] In the preferred embodiment of the above-mentioned method and system for collecting scientific and technological intelligence based on multimodal data acquisition, the specific steps for calculating the intelligence timeliness index TI are as follows: Basic intelligence data include intelligence release time t, decay rate ki, obsolescence threshold vt and maximum effective period Tmax; The intelligence timeliness index is calculated based on the intelligence release time t, decay rate ki, expiration threshold vt and maximum effective period Tmax. , the specific formula is as follows: .

[0008] In the preferred embodiment of the above-mentioned method and system for collecting scientific and technological intelligence based on multimodal data acquisition, the specific method for screening and collecting basic intelligence within the time limit is: When the intelligence timeliness index TI ≤ the intelligence timeliness index threshold F, it means that the basic intelligence is within the timeliness period; When the intelligence timeliness index TI>intelligence timeliness index threshold F, it means that the basic intelligence is out of time.

[0009] In the preferred embodiment of the above-mentioned method and system for collecting scientific and technological intelligence based on multimodal data acquisition, the specific steps of calculating the text index dex are as follows: Text data includes the reliability score Sou, text integrity score Key, text content relevance score Con, text length len, and maximum text length len of some basic intelligence belonging to text data. max , minimum text length len min , text publishing time Ti and average text publishing time Ti ang ; The specific formula for calculating the text index dex of different basic intelligence based on text data is as follows: , Among them, α represents the weight coefficient of the source reliability score Sou, with a value range of 0.1 < α < 0.4; β represents the weight coefficient of the text integrity score Key, with a value range of 0.2 < β < 0.5; γ represents the weight coefficient of the content relevance score Con, with a value range of 0.3 < γ < 0.4, and α + β + γ = 1; f represents the weight coefficient of, with a value range of 0.1 < f < 1.

[0010] In the above preferred solution of the scientific and technological intelligence collection method and system based on multi-modal data acquisition: The specific steps for calculating the image index ma are as follows: The image data includes the image reliability score cla, the image integrity score res, the image content relevance score tec, the minimum value of the resolution , the maximum value , the shooting time and the average shooting time ; According to the image data, calculate the image index ma of different basic information. The specific formula is as follows: , Among them, x1 represents the weight coefficient of, with a value range of 0.2 < x1 < 0.5, x2 represents the weight coefficient of, with a value range of 0.1 < x2 < 0.4, x3 represents the image content relevance score tec, with a value range of 0.3 < x3 < 0.4, and x1 + x2 + x3 = 1.

[0011] In the above preferred solution of the scientific and technological intelligence collection method and system based on multi-modal data acquisition: The specific steps for calculating the average shooting time are as follows: The image data also includes the number of images ; According to the shooting time and the number of images , calculate the average shooting time . The specific formula is as follows: , Among them, represents the shooting time of the j-th image, j represents the ordinal number of the image, and the value range is .

[0012] In the above preferred solution of the scientific and technological intelligence collection method and system based on multi-modal data acquisition: The specific steps for calculating the audio index tex are as follows: The audio data includes the sound quality score sou, the audio integrity score cty, and the audio technology relevance score nce of several basic information belonging to the audio data; The audio index tex of different basic information is calculated according to the audio data, and the specific formula is as follows: 。

[0013] In the above preferred solution of the scientific and technological intelligence collection method and system based on multi-modal data: The specific steps for calculating the video index vid are as follows: The video data includes the video reliability score vis, the video integrity score yj, and the video content relevance score toi of several basic information belonging to the video data; The video index vid of different basic information is calculated according to the video data, and the specific formula is as follows: , where, w1 represents the weight coefficient of the video reliability score with a value range of 0.1 < w1 < 0.4; w2 represents the weight coefficient of the video integrity score with a value range of 0.2 < w2 < 0.5; w3 represents the weight coefficient of the video content relevance score with a value range of 0.3 < w3 < 0.4, and w1 + w2 + w3 = 1.

[0014] In the above preferred solution of the scientific and technological intelligence collection method and system based on multi-modal data: The method for screening basic information that meets the standards is as follows: The basic information threshold set includes the text index threshold JV, the image index threshold GV, the audio index threshold TJ, and the video index threshold VJ; When the text index dex < the text index threshold JV, it means that this basic information does not meet the text data collection standard, and the text data needs to be recollected for calculation; when the text index dex ≥ the text index threshold JV, it means that this basic information meets the text data collection standard, and this basic information is retained; When the image index ma < the image index threshold GV, it means that this basic information does not meet the image data collection standard, and the image data needs to be recollected for calculation; when the image index ma ≥ the image index threshold GV, it means that this basic information meets the image data collection standard, and this basic information is retained; When the audio index tex < the audio index threshold TJ, it means that this basic information does not meet the audio data collection standard, and the audio data needs to be recollected for calculation; when the audio index tex ≥ the audio index threshold TJ, it means that this basic information meets the audio data collection standard, and this basic information is retained; When the video index vid is less than the video index threshold VJ, it means that the basic intelligence does not meet the video data collection standard, and the video data is collected again for calculation; when the video index vid is greater than or equal to the video index threshold VJ, it means that the basic intelligence meets the video data collection standard, and the basic intelligence is retained.

[0015] The present invention also discloses a method and system for collecting scientific and technological intelligence based on multimodal data acquisition, including: The data collection module is used to collect basic intelligence data, which includes some basic intelligence; The data analysis module is used to calculate the intelligence timeliness index TI of a number of basic intelligences based on the intelligence basic data; set the intelligence timeliness index threshold F, and screen and collect the basic intelligence within the timeliness according to the comparison result between the intelligence timeliness index TI and the intelligence timeliness index threshold F, and classify the screened basic intelligence into text data, image data, audio data and video data; Data sorting module, used to calculate the text index dex based on text data, the image index ma and the average shooting time based on image data , calculate the audio index tex according to the audio data, and calculate the video index vid according to the video data; The data screening module is used to set the basic intelligence threshold set, compare the text index dex, image index ma, audio index tex and video index vid with the corresponding thresholds in the basic intelligence threshold set, and screen the basic intelligence that meets the standards.

[0016] (III) Beneficial effects The present invention provides a method and system for collecting scientific and technological intelligence based on multimodal data acquisition, which has the following beneficial effects: (1) Collect basic intelligence data and gather original materials to provide rich resource support for intelligence collection, lay the foundation for accurate screening and in-depth analysis, and start a comprehensive and systematic intelligence mining process.

[0017] (2) Calculate and screen time-sensitive intelligence. Based on the intelligence, calculate and compare it with the threshold, accurately screen out the time-sensitive intelligence and classify it into multimodal data. This not only ensures the timeliness of the intelligence, but also diverts the fine processing of each mode, thereby improving the targeted processing.

[0018] (3) Through reliability score sou, text completeness score key, text content relevance score con, text length len, and maximum text length len max , minimum text length len min , text publishing time Ti and average text publishing time Ti angThe text index dex is calculated, which can deeply mine the scientific and technological intelligence in the text, avoid the limitations of simple keyword search, more accurately extract valuable scientific and technological intelligence content, and improve the accuracy and comprehensiveness of text intelligence analysis; through the image reliability score cla, image integrity score res, image content relevance score tec, and the minimum value of resolution , maximum value , Shooting time And the average shooting time , calculate the image index ma of different basic intelligence, greatly improve the accuracy and efficiency of recognition, and can more comprehensively obtain scientific and technological intelligence from image data; use sound quality score sou, audio integrity score and audio technology relevance score , the audio index tex is obtained, which can dig out the deep-level science and technology related information in the audio, rather than being limited to simple basic parameter measurements, thereby improving the ability to obtain valuable scientific and technological intelligence from audio data. The video reliability score vis, the video integrity score yj and the video content relevance score toi are comprehensively considered, and the video index vid is obtained through comprehensive quantification, which avoids the problem of only roughly extracting part of the picture for analysis and processing the audio and video separately. It can comprehensively and accurately extract the scientific and technological intelligence in the video, and improve the quality of video data intelligence analysis.

[0019] (4) By comparing the text index dex with the text index threshold JV, text data that meets the standards can be screened out, preventing low-quality text from interfering with subsequent analysis, thereby improving processing accuracy and efficiency; by comparing the image index ma with the image index threshold GV, image intelligence that meets the standards can be identified, and poor-quality and irrelevant images can be excluded, thereby ensuring the reliability and effectiveness of image analysis; by comparing the audio index tex with the audio index threshold TJ, qualified audio data can be screened out, avoiding analysis errors caused by poor audio quality, thereby improving the accuracy of audio intelligence processing; by comparing the video index vid with the video index threshold VJ, video data that meets the standards can be screened out, preventing low-quality video from affecting analysis results, thereby ensuring the reliability of video intelligence processing. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 It is a flow chart of the scientific and technological intelligence collection method based on multimodal data acquisition of the present invention. DETAILED DESCRIPTION

[0021] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in 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. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0022] See also Figure 1 The present invention provides a method and system for collecting scientific and technological intelligence based on multimodal data acquisition, including: Step 1: Collect basic intelligence data, which includes some basic intelligence.

[0023] Step 2: Based on the basic intelligence data, calculate the intelligence timeliness index TI of several basic intelligences; set the intelligence timeliness index threshold F, and based on the comparison result between the intelligence timeliness index TI and the intelligence timeliness index threshold F, screen and collect the basic intelligence within the timeliness, and classify the screened basic intelligence into text data, image data, audio data and video data.

[0024] Step 201, the specific steps of calculating the intelligence timeliness index TI are as follows: Basic intelligence data include intelligence release time t, decay rate ki, obsolescence threshold vt and maximum effective period Tmax.

[0025] It should be noted that the intelligence release time t can usually be calculated by subtracting the intelligence release time from the current time.

[0026] The decay rate ki is obtained by analyzing historical intelligence data, collecting multiple sets of intelligence value scores and corresponding time data , tn represents different time points, Represents the value scores at different time points. The value scores can be obtained through professional intelligence analysis tools, such as Yuanting Technology's defense intelligence classification management system, K2 visual analysis and judgment software, Huaxun intelligence analysis and judgment system, etc., to obtain the value scores at different time points. Take the natural logarithm , using the linear regression method, according to the principle of least squares, fitting the straight line , for linear regression, the calculation formula is: , in, represents the nth time point, Represents the value of intelligence at the nth time point.

[0027] In this situation , , , .

[0028] The obsolete threshold vt, when the value of intelligence is reduced to a certain proportion of the original value, the value of the intelligence is low and can be considered obsolete. For example, when the value of intelligence is reduced to 30% of the original value, the value of the intelligence is low, so the obsolete threshold vt is set to 30%.

[0029] The maximum effective period Tmax is calculated by analyzing the intelligence and calculating the average value of the effective period supported by various types of intelligence after their release, which is set as the maximum effective period Tmax.

[0030] The intelligence timeliness index is calculated based on the intelligence release time t, decay rate ki, expiration threshold vt and maximum effective period Tmax. , the specific formula is as follows: .

[0031] It should be noted that the attenuation part , the exponential function indicates that the value of intelligence decays at a certain rate over time. Threshold part , which partly reflects the degree to which the intelligence is close to being outdated. near When the value approaches 0, it means the intelligence is close to being outdated. Multiply the above two parts to get TI. When TI is less than or equal to the intelligence timeliness index threshold F, the intelligence is within the timeliness period; when TI is greater than F, the intelligence exceeds the timeliness period. Through this calculation, the timeliness of intelligence can be quantitatively evaluated.

[0032] By introducing parameters such as intelligence release time t, decay rate ki, expiration threshold vt and maximum effective period Tmax, the timeliness of intelligence can be quantified comprehensively and scientifically. Some of them can reflect the decay of intelligence value over time, while others take into account the validity period of intelligence. By comparing the calculated TI with the threshold F, it is possible to accurately determine whether the intelligence is within the validity period, avoiding the subjectivity and uncertainty of manual judgment, and helping to efficiently screen out intelligence with time-limited value, improve the efficiency and accuracy of intelligence processing, and ensure the quality of the intelligence used.

[0033] Step 202, the specific method of screening the basic intelligence collected within the time limit is: When the intelligence timeliness index TI ≤ the intelligence timeliness index threshold F, the intelligence is within the timeliness period.

[0034] When the intelligence timeliness index TI> the intelligence timeliness index threshold F, the intelligence exceeds the timeliness.

[0035] Step 3: Calculate the text index dex based on the text data, and calculate the image index ma and the average shooting time based on the image data , calculate the audio index tex based on the audio data, and calculate the video index vid based on the video data.

[0036] Step 301: The specific steps of calculating the text index dex are: Text data includes the reliability score Sou, text integrity score Key, text content relevance score Con, text length len, and maximum text length len of some basic intelligence belonging to text data. max , minimum text length len min , text publishing time Ti and average text publishing time Ti ang .

[0037] It should be noted that the reliability score Sou of basic intelligence is obtained by counting the number of words in the text that belong to the keyword set and then dividing it by the total vocabulary of the text to get the percentage of keyword appearance. Assuming the total vocabulary of the text is N and the number of keyword appearances is ni, the keyword appearance percentage P=(ni / N)×100, and the percentage P is mapped to a score of 0-100. For example, if P=30%, then sou=30.

[0038] The text integrity score Key is calculated by counting the number of words in the text that describe the technical principles and dividing it by the total number of words in the text to get the proportion of technical principles. Assuming that the number of words in the technical principle part is L1 and the total number of words in the text is L, then the proportion of technical principles is P1=(L1 / L)×100. For the proportion of experimental data, similarly, the number of words in the experimental data description part is L2, and the proportion of experimental data is P2=(L2 / L)×100. For the proportion of application scenarios, the number of words in the application scenario description part is L3, and the proportion of application scenarios is P3=(L3 / L)×100. The sum of these three proportions is used as the text integrity score key=P1+P2+P3.

[0039] The text content relevance score Con converts the text to be evaluated and the target topic text into vectors through the bag-of-words model. The bag-of-words model simply counts the number of times each word appears in the text to form a vector; the word embedding model maps the word to a low-dimensional vector space, which can capture the semantic information of the word. Assume that the text vector to be evaluated is =(a1,a2,...,an), the target topic text vector is =(b1,b2,...,bn), then the cosine similarity The formula is: , Map the cosine similarity value to a score from 0 to 100. For example, if the cosine similarity is 0.7, then con = 70.

[0040] The text length len can usually be obtained by simply counting the number of characters, words, or sentences in the text. The maximum text length len max and the minimum text length len min The acquisition method is to count the text length len of all basic intelligence, and obtain the maximum and minimum values ​​of the text length len as the maximum text length len max and the minimum text length len min .

[0041] The text publishing time Ti can usually be obtained from the original data of the text. For example, in a webpage article, there may be a label of the publishing time; in a document file, there may be a record of the creation time or modification time.

[0042] Average text publishing time Ti ang You need to first obtain the release time Ti of all texts, then calculate the average of these release times to get the average text release time Ti ang。

[0043] The specific formula for calculating the text index dex of different basic intelligence based on text data is as follows: , Among them, α represents the weight coefficient of the source reliability score Sou, and its value is 0.1<α<0.4; β represents the weight coefficient of the text integrity score Key, and its value is 0.2<β<0.5; γ represents the weight coefficient of the content relevance score Con, and its value is 0.3<γ<0.4, and α+β+γ=1; f represents The weight coefficient is 0.1 <f<1。

[0044] It should be noted that This part takes into account three main characteristics of the text: This part takes into account the effect of text length. This part takes into account the impact of the time when the text was published. By multiplying these three parts, we finally get the text index, which comprehensively considers factors such as the reliability of the text's source, keyword density, content relevance, text length and publication time.

[0045] The formula takes multiple factors into consideration. It combines source reliability, keyword density, content relevance, text length and publication time to comprehensively evaluate the value of the text and avoid the one-sidedness of single-factor evaluation. The weights are set reasonably and the importance of each factor is balanced through different weight coefficients. It can be flexibly adjusted according to actual needs to ensure that key factors have a reasonable impact on the results. The text length is normalized so that texts of different lengths are treated fairly in the evaluation. The processing of publication time can also reflect the impact of the text timeliness on its value, which helps to screen out more valuable and timely texts.

[0046] Step 302: The specific steps of calculating the image index ma are: The image data includes the image reliability score cla, image integrity score res, image content relevance score tec, and the minimum value of resolution, which are some basic intelligence of the image data. , maximum value , Shooting time And the average shooting time .

[0047] It should be noted that the image reliability score cla is obtained by classifying and scoring the image sources. Officially released images are scored at 0.8-1.0 points, and images uploaded by users and not officially verified are scored at 0.3-0.5 points. The image integrity score res is obtained by Adobe Photoshop, which extracts image features, identifies key image elements, and counts the number of key image elements that appear. For example, in military base intelligence images, key image elements include military facilities and weapons and equipment, defense facilities, the number of personnel and active soldiers, and the base layout and infrastructure building structure. The identification and quantitative statistics of these key image elements can provide important data support for intelligence analysis, help make accurate military assessments and decisions, and then calculate the percentage of key image elements that appear, which is the image integrity score percentage res.

[0048] The image content relevance score extracts the feature vectors of an image using image content analysis technology through tecAdobePhotoshop, compares the feature vectors with those of images related to the target theme, and determines the relevance score tec by calculating the similarity. For example, first, determine the target theme, such as "beach scenery", and collect high-quality reference images related to it. Then, open the image to be evaluated for relevance in AdobePhotoshop, and convert it if it is not in digital format. Use the built-in functions of AdobePhotoshop for image content analysis, extract features such as the color histogram, texture, and shape of the image and combine them into feature vectors. At the same time, extract the feature vectors of the reference images using the same method. Then, select a suitable similarity measurement method, such as cosine similarity, etc., and compare the feature vectors of the image to be evaluated with those of the reference images one by one to calculate the similarity. Finally, set the scoring criteria. For example, when the similarity is between 0.8 - 1.0, the relevance score is 90 - 100 points; when the similarity is between 0.6 - 0.8, the score is 70 - 90 points, etc. Determine the relevance score of the image according to the calculated similarity and the set criteria, and complete the evaluation of the image content relevance.

[0049] The minimum value of the resolution , the maximum value Record the minimum value of the image resolution through an image viewing software, such as IrfanView, etc. and the maximum value .

[0050] Calculate the average shooting time The specific steps are as follows: The image data also includes the number of images .

[0051] According to the shooting time and the number of images , calculate the average shooting time , and the specific formula is as follows: , where, represents the time of the j-th image, j represents the ordinal number of the image, and the value range is .

[0052] Calculate the image index ma of different basic information according to the image data, and the specific formula is as follows: where, x1 represents the weight coefficient of, and the value range is 0.2 < x1 < 0.5, x2 represents the weight coefficient of, and the value range is 0.1 < x2 < 0.4, x3 represents the image content relevance score The weight coefficient has a value range of 0.3 < x3 < 0.4, and x1 + x2 + x3 = 1.

[0053] It should be noted that This part takes into account the influence of image reliability score and resolution. This part takes into account the influence of image integrity score and shooting time. This part takes into account the relationship between shooting time and average shooting time. This part takes into account the influence of image content relevance score. By adding these four parts together, the image index is finally obtained, comprehensively considering factors such as image reliability, integrity, content relevance, resolution, and shooting time.

[0054] This formula combines multiple factors such as image reliability, integrity, content relevance, resolution, and shooting time, avoiding the one-sidedness of single-factor evaluation, being able to comprehensively reflect the value of the image, balancing the importance of each factor through different weight coefficients, being flexibly adjustable according to actual needs, ensuring that key factors have a reasonable impact on the result, and scientifically processing shooting time and resolution. It takes into account the relationship between shooting time and average shooting time, as well as the normalization processing of resolution, can more accurately evaluate the value of the image in terms of time and resolution, and helps to screen out high-quality images that meet the requirements.

[0055] Step 303: The specific steps for calculating the audio index tex are as follows: The audio data includes the sound quality score sou, audio integrity score cty, and audio technology relevance score nce of several basic information belonging to the audio data.

[0056] It should be noted that the sound quality score sou is obtained from the Adobe Audition audio processing software; the audio integrity score cty is obtained from the CoolEditPro audio integrity processing software; the audio technology relevance score nce is obtained through the IBM Watson Speech to Text audio processing software. Specifically as follows: Select the audio file to be analyzed in the software interface, and the audio can be preprocessed according to the situation. Then start the analysis. The software will perform speech recognition and transcription on the audio. After the transcription is completed, start the technical relevance analysis. On the one hand, search for audio technology-related keywords in the transcribed text, such as "audio encoding", etc. On the other hand, perform semantic analysis to determine the depth and breadth of the text in the audio technology field. After that, perform score calculation, assign weights to the identified keywords according to the preset keyword weight table, and at the same time score the depth of the text in the audio technology field based on the semantic analysis result. Finally, combine the keyword weight score and the semantic depth score to obtain the audio technology relevance score nce.

[0057] The specific formula for calculating the audio index tex of different basic intelligence based on audio data is as follows: .

[0058] It should be noted that It means adding the sound quality score, completeness score and technical relevance score respectively, and then dividing by 3 to get the average of the three scores of the three audios. The final result of the formula is to add the above two parts to get the audio index tex. The first part is the average of the total of the three scores, and the second part is the proportional value related to the product and sum of the three scores, which comprehensively considers the sound quality, integrity and technical relevance of the audio.

[0059] The formula combines the audio's sound quality score, integrity score, and technical relevance score, avoiding the one-sidedness of relying on a single factor to evaluate audio. It can comprehensively reflect the quality of the audio, and through summing and averaging, the comprehensive performance of different audios on these three indicators can be quantitatively compared, which helps to screen out audio with better overall performance. The second half of the formula further refines the evaluation mechanism through the proportional relationship between the product and sum of the three scores, taking into account the relationship between each score, and can more accurately reflect the actual value of the audio in multiple dimensions.

[0060] Step 304: The specific steps of calculating the video index vid are: The video data includes a video reliability score vis, a video integrity score yj, and a video content relevance score toi, which are some basic intelligence belonging to the video data.

[0061] It should be noted that the video reliability score vis can use existing video detection tools, such as Tencent Cloud Media Quality Inspection Software, to check whether the video covers 13 detection types such as screen distortion, black edges, mosaics, noise, etc., and provide an overall video quality detection score. The video reliability score vis is obtained by summing up the detection and scoring of various aspects of the video quality.

[0062] Video integrity scoring can be done by using video analysis tools such as Yuanchuangbao and Douyin and Kuaishou short video analysis to detect and analyze non-original elements, and determine whether the video has problems that affect integrity, such as missing or replaced content. A complete video with no missing content is scored 100 points, and a video with suspected missing content is scored 50 points. The video content relevance score toi is obtained through the IBM Watson Media video content analysis software. According to the target theme, the extracted keywords and the identified content are matched with the target theme. In terms of keyword matching, the frequency and importance of keywords related to the target theme appearing in the video are calculated. For example, for a theme about "the application of artificial intelligence in medicine", the occurrences of keywords such as "medical image diagnosis" and "machine learning algorithms" in the video are counted. In terms of semantic analysis, by comprehensively understanding the semantics of the video content, the relevance of the video content to the target theme is judged. For example, it is analyzed whether the video content is discussing how artificial intelligence helps doctors in disease diagnosis and other related content. Based on the results of keyword matching and semantic analysis, the video content relevance score toi is comprehensively calculated.

[0063] The video index vid of different basic information is calculated based on the video data, and the specific formula is as follows: , where w1 represents the weight coefficient of the video reliability score with a value range of 0.1 < w1 < 0.4; w2 represents the weight coefficient of the video integrity score with a value range of 0.2 < w2 < 0.5; w3 represents the weight coefficient of the video content relevance score with a value range of 0.3 < w3 < 0.4, and w1 + w2 + w3 = 1.

[0064] It should be noted that the formula comprehensively evaluates the quality of the video by performing weighted summation on these three aspects of reliability, integrity, and relevance.

[0065] The reliability score vis multiplied by its weight coefficient w1 reflects the contribution of the video source reliability to the video index. The integrity score yj multiplied by its weight coefficient w2 reflects the impact of the degree of key information contained in the video on the video index. The relevance score toi multiplied by its weight coefficient w3 represents the role of the video content's relevance to the target theme in the video index. Finally, these three parts are added together to obtain the video index vid. The higher its value, the higher the comprehensive quality of the video.

[0066] By comprehensively considering the video reliability score vis, the integrity score yj, and the content relevance score toi, and setting reasonable weight coefficients, this formula can comprehensively and scientifically evaluate the video quality, avoiding the one-sidedness of single-factor evaluation. Reasonable value range restrictions are imposed on the weight coefficients to ensure that the proportion of each factor in the evaluation is relatively scientific, making the evaluation results more accurate and reliable. This quantitative calculation method can conveniently screen and sort a large number of videos, and can efficiently identify high-quality videos during the information processing process, improving work efficiency and the accuracy of information processing.

[0067] Step 4: Set a basic intelligence threshold set, compare the text index dex, image index ma, audio index tex, and video index vid with the corresponding thresholds in the basic intelligence threshold set, and screen the basic intelligence that meets the standards.

[0068] The method of selecting basic intelligence that meets the criteria is: The text index threshold JV is obtained by collecting text data samples such as official texts or documents. These samples should cover texts of different types, fields and quality levels. For example, academic papers, news reports, technical documents, etc. For each text sample, its text index dex is calculated, and then the text index dex values ​​of all text samples are summarized to calculate the average value of these text index dex values. This average value can be used as the standard for the text index threshold JV.

[0069] The image index threshold GV is obtained by collecting official image data samples, which should cover images of different types, fields and quality levels. For example, scientific research instruments, technical products, experimental scenes, etc. These samples should include different types of images, such as landscape photos, portrait photos, product photos, works of art, etc., and the image quality is high or low. For each image sample, its image index ma is calculated, and the image index ma values ​​of all image samples are summarized. The average value of these image index ma values ​​is calculated, and the average value is used as the standard for the image index threshold GV.

[0070] The audio index threshold TJ is calculated by collecting official audio data samples. These samples can cover audio of different types, different fields and different quality levels, such as the live sound of technology product launches, the sound generated by the operation of scientific research instruments, recordings of technology lectures, etc., including music, voice, environmental sound effects, etc., and the audio quality varies. For each audio sample, its audio index tex is calculated, and the audio index tex values ​​of all audio samples are summarized, and the average of these audio index tex values ​​is calculated. This average value is used as the standard for the audio index threshold TJ.

[0071] Video index threshold VJ, by collecting official video data samples, these samples can cover videos of different types, different fields and different quality levels, such as movie clips, documentaries, advertisements, home videos, etc., the video quality varies. For each video sample, its video index vid is calculated, the video index vid values ​​of all video samples are summarized, and the average of these video index vid values ​​is calculated. The average value is used as the standard for the video index threshold VJ.

[0072] When the text index dex is less than the text index threshold JV, it means that the basic intelligence does not meet the text data collection standard, and the text data is collected again for calculation; when the text index dex is greater than or equal to the text index threshold JV, it means that the basic intelligence meets the text data collection standard, and the basic intelligence is retained.

[0073] When the image index ma is less than the image index threshold GV, it indicates that the basic intelligence does not meet the image data collection standard, and the image data is collected again for calculation; when the image index ma is greater than or equal to the image index threshold GV, it indicates that the basic intelligence meets the image data collection standard, and the basic intelligence is retained.

[0074] When the audio index tex is less than the audio index threshold TJ, it means that the basic intelligence does not meet the audio data collection standard, and the audio data is collected again for calculation; when the audio index tex is greater than or equal to the audio index threshold TJ, it means that the basic intelligence meets the audio data collection standard, and the basic intelligence is retained.

[0075] When the video index vid is less than the video index threshold VJ, it means that the basic intelligence does not meet the video data collection standard, and the video data is collected again for calculation; when the video index vid is greater than or equal to the video index threshold VJ, it means that the basic intelligence meets the video data collection standard, and the basic intelligence is retained.

[0076] It should be noted that the whole process is a step-by-step screening process, in which the correct data is screened out by comparing the text index, image index, audio index and video index with the threshold. This method decomposes complex multimodal data processing into multiple single-modal data processing steps, which is easy to operate and manage.

[0077] By filtering data by comparing with the threshold, data of poor quality can be effectively eliminated. This sub-modal and sub-step screening method helps to optimize the data processing process. It enables data processing personnel to process data of different modalities in a targeted manner and improve processing efficiency. At the same time, by setting thresholds for screening, the subjectivity of manual judgment is reduced, making the data screening process more objective and scientific, and ensuring data consistency. In multimodal data processing, this screening method helps to ensure consistency between data of different modalities. Only when each modal data passes its own screening criteria will it enter the next step of processing, avoiding contradictions and mismatches between data of different modalities, thereby ensuring data consistency and availability.

[0078] On the other hand, the present invention also discloses a method and system for collecting scientific and technological intelligence based on multimodal data acquisition, including: The data collection module is used to collect basic intelligence data, which includes some basic intelligence.

[0079] The data analysis module is used to calculate the intelligence timeliness index TI of several basic intelligences based on the basic intelligence data; set the intelligence timeliness index threshold F, and screen and collect the basic intelligence within the timeliness based on the comparison result between the intelligence timeliness index TI and the intelligence timeliness index threshold F, and classify the screened basic intelligence into text data, image data, audio data and video data.

[0080] Data sorting module, used to calculate the text index dex based on text data, the image index ma and the average shooting time based on image data , calculate the audio index tex based on the audio data, and calculate the video index vid based on the video data.

[0081] The data screening module is used to set the basic intelligence threshold set, compare the text index dex, image index ma, audio index tex and video index vid with the corresponding thresholds in the basic intelligence threshold set, and screen the basic intelligence that meets the standards.

[0082] The above embodiments may be implemented in whole or in part by software, hardware, firmware or any other combination thereof. When implemented using software, the above embodiments may be implemented in whole or in part in the form of a computer program product. A person of ordinary skill in the art may appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein may be implemented in electronic hardware or in combination with computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution.

[0083] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0084] The above description is only a specific implementation manner of the present application, but the protection scope of the present application is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application.

Claims

1. A method for collecting scientific and technological intelligence based on multimodal data acquisition, characterized in that: include: Collect basic intelligence data, which includes some basic intelligence; Calculate the intelligence timeliness index TI of some basic intelligence based on basic intelligence data; Set an intelligence timeliness index threshold F, and based on the comparison result between the intelligence timeliness index TI and the intelligence timeliness index threshold F, screen and collect basic intelligence within the timeliness, and classify the screened basic intelligence into text data, image data, audio data, and video data; The text index dex is calculated based on the text data, and the image index ma and the average shooting time are calculated based on the image data. , calculate the audio index tex according to the audio data, and calculate the video index vid according to the video data; A basic intelligence threshold set is set, and the text index dex, image index ma, audio index tex, and video index vid are compared with the corresponding thresholds in the basic intelligence threshold set to screen basic intelligence that meets the standards.

2. The method for collecting scientific and technological intelligence based on multimodal data acquisition according to claim 1, characterized in that: The specific steps for calculating the intelligence timeliness index TI are: Basic intelligence data include intelligence release time t, decay rate ki, obsolescence threshold vt and maximum effective period Tmax; The intelligence timeliness index is calculated based on the intelligence release time t, decay rate ki, expiration threshold vt and maximum effective period Tmax. The specific formula is as follows: 。 3. The method for collecting scientific and technological intelligence based on multimodal data acquisition according to claim 2 is characterized in that: The specific method for screening and collecting basic intelligence within the time limit is: When the intelligence timeliness index TI ≤ the intelligence timeliness index threshold F, it means that the basic intelligence is within the timeliness period; When the intelligence timeliness index TI>intelligence timeliness index threshold F, it means that the basic intelligence is out of time.

4. The method for collecting scientific and technological intelligence based on multimodal data acquisition according to claim 3 is characterized in that: The specific steps for calculating the text index dex are: Text data includes the reliability score Sou, text integrity score Key, text content relevance score Con, text length len, and maximum text length len of some basic intelligence belonging to text data. max , minimum text length len min , text publishing time Ti and average text publishing time Ti ang ; The specific formula for calculating the text index dex of different basic intelligence based on text data is as follows: ， Among them, α represents the weight coefficient of the source reliability score Sou, and its value is 0.1<α<0.4; β represents the weight coefficient of the text integrity score Key, and its value is 0.2<β<0.5; γ represents the weight coefficient of the content relevance score Con, and its value is 0.3<γ<0.4, and α+β+γ=1; f represents The weight coefficient is 0.1 <f<1。 5. The method for collecting scientific and technological intelligence based on multimodal data acquisition according to claim 4 is characterized in that: The specific steps for calculating the image index ma are: The image data includes the image reliability score cla, image integrity score res, image content relevance score tec, and the minimum value of resolution, which are some basic intelligence of the image data. , maximum value , Shooting time And the average shooting time ; The specific formula for calculating the image index ma of different basic intelligence based on image data is as follows: ， Among them, x1 represents the weight coefficient of, and the value range is 0.2 < x1 < 0.

5. x2 represents the weight coefficient of, and the value range is 0.1 < x2 < 0.

4. x3 represents the weight coefficient of, and the value range is 0.3 < x3 < 0.4, and x1 + x2 + x3 = 1.

6. The method for collecting scientific and technological intelligence based on multimodal data acquisition according to claim 5 is characterized in that: Calculate the average shooting time The specific steps are: Image data also includes the number of images ; According to shooting time And the number of images , calculate the average shooting time The specific formula is as follows: ， in, Represents the shooting time of the jth image, j represents the ordinal number of the image, and its value is .

7. The method for collecting scientific and technological intelligence based on multimodal data acquisition according to claim 6 is characterized in that: The specific steps for calculating the audio index tex are: The audio data includes the sound quality score sou, audio integrity score, and some basic information belonging to the audio data. and audio technology relevance score ; The specific formula for calculating the audio index tex of different basic intelligence based on audio data is as follows: 。 8. The method for collecting scientific and technological intelligence based on multimodal data acquisition according to claim 7, characterized in that: The specific steps for calculating the video index vid are: The video data includes a video reliability score vis, a video integrity score yj, and a video content relevance score toi, which are some basic intelligence of the video data; The specific formula for calculating the video index vid of different basic intelligence based on video data is as follows: ， Among them, w1 represents the weight coefficient of the video reliability score , with a value range of 0.1 < w1 < 0.4; w2 represents the weight coefficient of the video integrity score , with a value range of 0.2 < w2 < 0.5; w3 represents the weight coefficient of the video content relevance score , with a value range of 0.3 < w3 < 0.4, and w1 + w2 + w3 = 1.

9. The method for collecting scientific and technological intelligence based on multimodal data acquisition according to claim 8, characterized in that: The method of selecting basic intelligence that meets the criteria is: The basic intelligence threshold set includes a text index threshold JV, an image index threshold GV, an audio index threshold TJ, and a video index threshold VJ; When the text index dex is less than the text index threshold JV, it means that the basic intelligence does not meet the text data collection standard, and the text data is collected again for calculation; when the text index dex is greater than or equal to the text index threshold JV, it means that the basic intelligence meets the text data collection standard, and the basic intelligence is retained; When the image index ma is less than the image index threshold GV, it means that the basic intelligence does not meet the image data collection standard, and the image data is collected again for calculation; when the image index ma is greater than or equal to the image index threshold GV, it means that the basic intelligence meets the image data collection standard, and the basic intelligence is retained; When the audio index tex is less than the audio index threshold TJ, it means that the basic intelligence does not meet the audio data collection standard, and the audio data is collected again for calculation; when the audio index tex is greater than or equal to the audio index threshold TJ, it means that the basic intelligence meets the audio data collection standard, and the basic intelligence is retained; When the video index vid is less than the video index threshold VJ, it means that the basic intelligence does not meet the video data collection standard, and the video data is collected again for calculation; when the video index vid is greater than or equal to the video index threshold VJ, it means that the basic intelligence meets the video data collection standard, and the basic intelligence is retained.

10. A scientific and technological intelligence collection system based on multimodal data acquisition, characterized by: The data collection module is used to collect basic intelligence data, which includes some basic intelligence; The data analysis module is used to calculate the intelligence timeliness index TI of a number of basic intelligences based on the intelligence basic data; set the intelligence timeliness index threshold F, and screen and collect the basic intelligence within the timeliness according to the comparison result between the intelligence timeliness index TI and the intelligence timeliness index threshold F, and classify the screened basic intelligence into text data, image data, audio data and video data; Data sorting module, used to calculate the text index dex based on text data, the image index ma and the average shooting time based on image data , calculate the audio index tex according to the audio data, and calculate the video index vid according to the video data; The data screening module is used to set the basic intelligence threshold set, compare the text index dex, image index ma, audio index tex and video index vid with the corresponding thresholds in the basic intelligence threshold set, and screen the basic intelligence that meets the standards.

Images