Intelligent Information Analysis System for Advertising Videos

Abstract

The present invention provides an intelligent information analysis system for advertising videos. The system uses a modular analysis and processing method. Based on natural language analysis, on the basis of segmenting the video, it sequentially advances to the character recognition module, the screening module, and the evaluation module to extract high-frequency vocabulary and advertising evaluation data. On this basis, two-level feedback, namely rough adjustment feedback and fine adjustment feedback, is provided to the initial video cutting module. Under the two-level feedback, the video cutting is made more accurate, thereby ensuring that the evaluation results are more precise, and thus ensuring that the advertising evaluation is more intelligent in a technical manner.

Description

Technical Field

[0001] The present invention relates to an intelligent information analysis system for advertising videos. Background Art

[0002] Currently, in the advertising industry, the quality of materials directly affects the advertising effect. Especially for video materials, the text information therein (such as brand names, advertising slogans, etc.) is of great value for brand recognition, content understanding, audience interaction, and keyword analysis.

[0003] However, current technologies mainly rely on extracting subtitle files or generating text from audio, which cannot effectively analyze the text appearing in the video at the visual level. In addition, traditional OCR technologies are mainly applied to the text recognition of static images and cannot deeply extract and screen effective information. This makes it difficult for users to comprehensively, quantitatively understand the reactions and interests of the audience.

[0004] Therefore, there is an urgent need in this field for an accurate and quantitative technology that can achieve precise advertising evaluation. Summary of the Invention

[0005] The present invention provides an intelligent information analysis system for advertising videos, which effectively solves the above problems in the prior art.

[0006] Specifically, the present invention provides an intelligent information analysis system for advertising videos, which includes a video segmentation module, a text recognition module, a screening module, an evaluation module, and a comprehensive analysis module. The video segmentation module extracts one frame of advertising picture from an advertising video with a duration of T every period t, thus extracting a total of n advertising pictures. T > t, n = [T / t]. The text recognition module uses an OCR tool to recognize the advertising text in the n advertising pictures. The screening module cleans the advertising text and then extracts k high-frequency words from the advertising text. Set the rated number K of high-frequency words. If k < K, send a coarse adjustment feedback signal to the video segmentation module, and adjust the period t to the period t' = t × k 2 / K 2 , the advertising pictures are adjusted to n' = [T / t'], the evaluation module grabs the advertising evaluations of a total number A based on the n' advertising pictures. Set the advertising evaluation quantity threshold as A*. If A ≤ A*, send a fine adjustment feedback signal to the video segmentation module, and the video segmentation module updates the period t' to t'' = t' × A / A*, and the video cutting module cuts out n'' advertising pictures, n'' = [T / t'']. The screening module extracts k'' high-frequency words for the n'' advertising pictures. The comprehensive analysis module comprehensively and quantitatively determines the comprehensive vocabulary evaluation index obtained by each high-frequency word based on the k'' high-frequency words and the advertising evaluations with a total number of A.

[0007] In particular, the advertisement text includes the brand name of the advertisement.

[0008] Preferably, the cleaning of the advertisement text by the screening module includes cleaning special characters and meaningless symbols in the advertisement text, unifying the case of letters in the advertisement text, and eliminating stop words.

[0009] Preferably, in the screening module, if k≥K, the time period t remains unchanged, t’=t, and correspondingly, the number of advertisement images also remains unchanged, so n’=n.

[0010] Preferably, the screening module sets a threshold number of times, and words in the advertisement text that exceed the threshold number of times are determined by the screening module as high-frequency words.

[0011] Preferably, in the evaluation module, all advertisement evaluations of all n’ frames corresponding to the n’ advertisement images are extracted, and the number of advertisement evaluations for each frame in the n’ advertisement images is a i , where the integer i ranges from 1 to n’, so the total number of advertisement evaluations of the n’ advertisement images is .

[0012] Preferably, in the evaluation module, if A>A*, the time period t’ remains unchanged, t’’=t’, and correspondingly, the number of advertisement images n’’=n’.

[0013] Preferably, in the comprehensive analysis module, the advertisement evaluations with a total number of A are quantitatively converted into A evaluation indexes one by one through natural language processing. The evaluation indexes are divided into positive evaluation indexes and negative evaluation indexes. The positive evaluation indexes are quantitatively converted into positive numbers, and the negative evaluation indexes are quantitatively converted into negative numbers.

[0014] Preferably, in the comprehensive quantitative determination, the number of any one high-frequency word extracted from the j-th advertisement image among the n’’ advertisement images is counted as b j , where the integer j ranges from 1 to n’’, and the quantitative evaluation index of the advertisement evaluation of the frame where the j-th advertisement image is located is B j , then the word evaluation index of any one high-frequency word in the j-th advertisement image is m j =B j / b j , if the j-th advertisement image does not appear any of the high-frequency words, then the word evaluation index m of any one high-frequency word in this advertisement image j is counted as 0. As the integer j ranges from 1 to n’’, the comprehensive word evaluation index of any one high-frequency word is calculated. .

[0015] Generally speaking, the present invention provides an intelligent information analysis system for advertising videos. This system adopts a modular analysis and processing method. Based on natural language analysis, on the basis of segmenting the video, it sequentially advances to the text recognition module, the screening module, and the evaluation module to extract high-frequency vocabulary and advertising evaluation data. On this basis, two-level feedback, namely rough adjustment feedback and fine adjustment feedback, is sent to the initial video cutting module. Under the two-level feedback, the video cutting is made more accurate, thus ensuring that the evaluation results are more precise. Thereby, it ensures that the advertising evaluation is more intelligent in a technical way. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will discuss the drawings required for use in the description of the embodiments or the prior art. Obviously, the technical solutions described in conjunction with the drawings are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other embodiments and their drawings can be obtained based on the embodiments shown in these drawings.

[0017] Figure 1 The flowchart of the operation of the intelligent information analysis system for advertising videos according to the present invention is shown. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0018] The following will clearly and completely describe the technical solutions of the embodiments of the present invention in conjunction with the drawings. Obviously, the described embodiments are only some embodiments of the present invention, rather than all embodiments. Based on the embodiments described in the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts are within the scope of protection of the present invention.

[0019] Generally speaking, the present invention provides an intelligent information analysis system for advertising videos. First, the video segmentation module segments the video and extracts the images of the corresponding time frames. The text recognition module then uses OCR technology to recognize the text information of each frame of the image, including the brand name and other advertising-related information. The screening module uses natural language technology to perform semantic analysis on the recognized text information and screen out high-frequency vocabulary. At the same time, it potentially feeds back to the video segmentation module based on the number of high-frequency vocabulary. The evaluation module deeply analyzes the correlation between high-frequency vocabulary and audience evaluation and gives relevant evaluation results. At the same time, it potentially sends feedback to the video segmentation module to ensure that the evaluation results are more precise. Thus, it can be seen that in the system of the present invention, double-loop feedback is carried out, and the large and small feedbacks ensure that the overall analysis is extremely precise.

[0020] The following will be combined with Figure 1 The specific content of the intelligent information analysis system for advertising videos provided by the present invention will be introduced in detail. Figure 1Shows the operation flowchart of the intelligent information analysis system for advertising videos according to the present invention.

[0021] As Figure 1 shown, the system first starts the segmentation module to perform segmentation processing on an advertising video. Specifically, for an advertising video with a duration of T, one frame of the advertising picture is extracted from the video every period t, and thus n = [T / t] advertising pictures are extracted, where T > t. The symbol "[]" here means rounding. For example, for an advertising video with a duration T = 53 minutes, one frame of the advertising picture is extracted every 5 minutes, then a total of 10 (=[53 / 5]) frames of advertising pictures are extracted.

[0022] As will be mentioned below, for the original advertising video, the duration T is a fixed value, but the period t may be adjusted in two ways with subsequent feedback. The reason why the present invention needs to consider adjusting the period t is that if the period t is too long, the number of extracted pictures is small, and it will be difficult to feedback the complete content of the advertisement, and the analysis result will be too rough. On the contrary, if the period t is too short, the number of extracted pictures is too large, which will greatly increase the analysis volume, make the analysis process too complicated, and easily lead to too much garbage information in the analysis.

[0023] Then start the text recognition module to extract the high-frequency words in the n advertising pictures.

[0024] Specifically, the text recognition module uses OCR tools (such as Tesseract, EasyOCR, etc.) to recognize and extract the advertising text in the n advertising pictures.

[0025] These advertising texts contain all the text information in the picture, including the brand name and other advertisement-related information. The so-called advertisement-related information, for example, is the advertising slogan, the performance price of the advertised product, etc. The relevant code can be described as follows:

[0026] import pytesseract

[0027] from PIL import Image

[0028] # Load the image file

[0029] image_path = 'frame0.jpg'

[0030] image = Image.open(image_path)

[0031] # Perform OCR recognition

[0032] text = pytesseract.image_to_string(image)

[0033] print(text)

[0034] At this time, the screening module is started to perform semantic extraction on the advertisement text. First, the screening module cleans the data of the advertisement text. For example, it cleans special characters and meaningless symbols in the advertisement text, unifies the case of letters in the advertisement text, and eliminates stop words (such as words like "de", "shi", "zai", etc.). For example, the relevant code is as follows:

[0035] import re

[0036] from nltk.corpus import stopwords

[0037] # Assume the data is stored in a list

[0038] texts =

[0039] "How to use OCR to identify high-frequency words in a video?

[0040] "What is the application of OCR in image processing?"

[0041] Can video analysis extract high-frequency words?

[0042] "What impact do high-frequency words have on text analysis?]

[0043] # Download NLTK stop words

[0044] import nltk

[0045] nltk.download('stopwords") stop words= set(stopwords.words('chinese”)) # Use in cleaned texts = for text in texts:

[0046] # Clean the text

[0047] text = re.sub(r"[^\W\s]”,"”,text) # Remove punctuation marks

[0048] text =text.lower() # Convert to lowercase cleaned texts.append(text)

[0049] # Also remove stop words

[0050] tokenized texts =[]

[0051] for text in cleaned texts:

[0052] words = text.split() # Tokenize the text

[0053] filtered_words = [word for word in words if word not in tokenized_texts] tokenized_texts.extend(filtered_words) # Add to the total words

[0054] The screening module then extracts high-frequency words from the cleaned advertisement texts. Specifically, words that appear more than a threshold number of times in the advertisement texts can be set as high-frequency words, and vice versa.

[0055] The screening module then counts the extracted high-frequency words. Suppose there are k high-frequency words counted. Set the rated number of high-frequency words as K. If k < K, it indicates that the extraction of high-frequency words in the advertisement text is insufficient. At this time, the coarse adjustment feedback is activated, and the screening module sends a coarse adjustment feedback signal to the video segmentation module to adjust the time period t by a multiple.

[0056] For example, through the multiple adjustment, the time period t is adjusted to the time period t'. The specific calculation formula is: t' = t × k 2 / K 2 , correspondingly, the video cutting module cuts out n' advertisement frames, where n' = [T / t']. Through this multiple adjustment, the number of advertisement frames that the video segmentation module can cut out can be greatly increased. The updated n' advertisement frames are input into the text recognition module for the recognition of advertisement texts, thereby increasing the number of recognition samples and making the recognition of advertisement texts more accurate.

[0057] If k ≥ K, the time period t remains unchanged, t' = t, and correspondingly, the number of advertisement frames also remains unchanged, so n' = n.

[0058] It should be noted that as mentioned above, the system of the present invention employs a dual-feedback mechanism. Therefore, the coarse adjustment feedback here is one of the feedback mechanisms, and another feedback mechanism, namely, the fine adjustment feedback, will be mentioned below.

[0059] Input the current n' advertisement frames and all high-frequency words into the evaluation module. It should be noted that when k ≥ K, the integer n' takes the value of the original number of advertisement frames n, that is, n' = n. When k < K, the integer n' takes the value of n' = [T / t'] = [T / (t × k 2 / K 2)]. In other words, according to the above different situations, the number n' of advertising screens in the evaluation module may have two different values.

[0060] The ultimate goal of the evaluation module is to give a final quantitative evaluation result for the advertising video. However, before giving the final quantitative evaluation result, the evaluation module also makes a fine-tuning feedback, which is the other feedback mechanism in the double-feedback mechanism mentioned above in addition to the rough-tuning feedback.

[0061] As is well known, the audience may evaluate the advertising video at any moment during the playback of the advertising video. In the evaluation module, all advertising evaluations of all n' frames corresponding to the n' advertising screens will be extracted (for example, by a crawler program), and the number of advertising evaluations for each frame in the n' advertising screens is a i , where the integer i takes values from 1 to n'. Therefore, the total number of advertising evaluations for the n' advertising screens is . Set the advertising evaluation quantity threshold as A*. If A ≤ A*, it indicates that the number of advertising evaluations is insufficient. In other words, the number of the n' advertising screens extracted is insufficient. Then the evaluation module will send a fine-tuning feedback signal to the video segmentation module, and the video segmentation module makes a multiple adjustment to the time period t' to update the time period t' to t'' = t' × A / A*. Correspondingly, the video cutting module cuts out n'' advertising screens, where n'' = [T / t''].

[0062] If A > A*, the time period t' remains unchanged, t'' = t'. Correspondingly, the number of advertising screens also remains unchanged, so n'' = n'. Through this multiple adjustment, the number of advertising screens that the video segmentation module can segment out can be slightly increased. The updated n'' advertising screens are input to the text recognition module for the recognition of advertising texts, so as to further increase the number of recognition samples through fine-tuning on the basis of rough-tuning, and also make the recognition of advertising texts more accurate. Correspondingly, the screening module extracts k'' high-frequency words for the n'' advertising screens.

[0063] Finally, the system starts the comprehensive analysis module to comprehensively and quantitatively determine the evaluation index obtained by each high-frequency word based on the k'' high-frequency words and the A advertising evaluations.

[0064] For example, the A advertisement evaluations can be quantitatively converted into A evaluation indices one by one through natural language processing. It should be noted that the evaluations of the audience are often divided into positive evaluations and negative evaluations. Positive evaluations are quantitatively converted into positive numbers, and negative evaluations are quantitatively converted into negative numbers. For example, if an advertisement evaluation is "I like this advertisement video very much", then the word "like" is extracted through natural language processing in this advertisement evaluation and quantitatively converted into "+2". Another advertisement is "This advertisement video is extremely wonderful", then the word "wonderful" is extracted through natural language processing in this advertisement evaluation and quantitatively converted into "+3"; on the contrary, it is a negative evaluation. For example, if an advertisement evaluation is "I hate this advertisement video", then the word "hate" is extracted through natural language processing and quantitatively converted into "-2".

[0065] In the above comprehensive quantitative determination, the number of any high-frequency word extracted from the j-th advertisement picture among the n'' advertisement pictures is counted as b j , and the quantitative evaluation index of the advertisement evaluation of the frame where the j-th advertisement picture is located is B j , then the word evaluation index of any high-frequency word in the j-th advertisement picture is m j = B j / b j . If the j-th advertisement picture does not appear any of the high-frequency words, the word evaluation index m j is counted as 0. As the integer j takes values from 1 to n'', the comprehensive word evaluation index of any high-frequency word can be calculated .

[0066] So far, the present invention has been basically introduced. Generally speaking, the present invention provides an intelligent information analysis system for advertisement videos. This system adopts a modular analysis and processing method. Based on natural language analysis, on the basis of video segmentation processing, it sequentially advances to the character recognition module, the screening module, and the evaluation module to extract high-frequency words and advertisement evaluation data. And on this basis, two-level feedback of rough adjustment feedback and fine adjustment feedback is carried out to the initial video cutting module. Under the two-level feedback, the video cutting is made more accurate, thereby ensuring that the evaluation result is more accurate. Thus, it ensures that the advertisement evaluation is more intelligent in a technical way.

[0067] The above are only exemplary embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. An advertising video intelligent information analysis system, characterized in that: The system includes a video segmentation module, a text recognition module, a screening module, an evaluation module, and a comprehensive analysis module, among which: The video segmentation module extracts one frame of advertising screen every time period t from an advertising video of duration T, thereby extracting a total of n advertising screens, T>t, n=[T / t], The text recognition module uses an OCR tool to recognize the advertisement texts in the n advertisement images. The screening module cleans the advertising text and then extracts k high-frequency words from the advertising text. Set the rated number K of high-frequency words. If k < K, send a rough adjustment feedback signal to the video segmentation module, and adjust the time period t to the time period t' = t × k 2 / K 2 , and adjust the advertising screen to n' = [T / t'], The evaluation module captures a total number A of advertising evaluations based on the n' advertising screens. Advertising evaluation refers to the evaluation of the advertising video by the audience at any time when the advertising video is played. The threshold of the number of advertising evaluations is set to A*. If A≤A*, a fine-tuning feedback signal is sent to the video segmentation module. The video segmentation module updates the time period t' to t"=t'ⅹA / A*. The video cutting module cuts out n" advertising screens, n"=[T / t"]. The return screening module extracts k" high-frequency words for the n" advertising screens. The comprehensive analysis module comprehensively and quantitatively determines the comprehensive vocabulary evaluation index obtained by each high-frequency vocabulary based on the k" high-frequency vocabulary and the total number of advertising evaluations A. In the comprehensive analysis module, the total number of advertising evaluations A is quantified into A evaluation indexes one by one through natural language processing. The evaluation index is divided into a positive evaluation index and a negative evaluation index. The positive evaluation index is quantified as a positive number, and the negative evaluation index is quantified as a negative number. In the comprehensive quantitative determination, the number of any high-frequency vocabulary extracted from any j-th advertising screen in the n" advertising screens is counted as b. j , where the integer j ranges from 1 to n", and the quantitative evaluation index of the advertisement evaluation of the frame where the j-th advertisement picture is located is B j , then the vocabulary evaluation index of any high-frequency word in the jth advertising screen is m j =B j / b j If any of the high-frequency words does not appear in the j-th advertising screen, the vocabulary evaluation index m of any of the high-frequency words in the advertising screen is j As the integer j changes from 1 to n", the comprehensive vocabulary evaluation index of any high-frequency word is calculated.

2. The system according to claim 1, characterized in that The advertisement text includes the brand name of the advertisement.

3. The system according to claim 1, characterized in that The screening module cleans the advertisement text by removing special characters and meaningless symbols in the advertisement text, unifying the case of letters in the advertisement text, and eliminating stop words.

4. The system according to claim 1, characterized in that In the screening module, if k≥K, the time period t remains unchanged, t'=t, and accordingly, the number of advertisement screens also remains unchanged, n'=n.

5. The system according to claim 1, characterized in that The screening module sets a threshold number of times, and the words in the advertisement text that exceed the threshold number of times are determined by the screening module as high-frequency words.

6. The system according to claim 1, characterized in that In the evaluation module, all advertisement evaluations of all n' frames corresponding to the n' advertisement pictures are extracted, and the number of advertisement evaluations of each frame in the n' advertisement pictures is a. i , where the integer i ranges from 1 to n', so the total number of advertising evaluations for the n' advertising screens is 7. The system according to claim 1, characterized in that In the evaluation module, if A>A*, the time period t' remains unchanged, t"=t', and accordingly, the number of advertisement screens n"=n'.

Images