Alibaba Cloud Big Data teaches you how to play in the entertainment industry

Not only that, but the public trend analysis turned out to be an eye of Alibaba small ai, helping small ai successfully predict the ranking of fans in the "I Am A Singer" finals. Today, the chef will take you to the entertainment circle with him.

Essential Products: http://click.aliyun.com/m/5647/

Unit Price: 69 RMB for the UI Public Edition/on August 2 on time!

The application scenarios of Alibaba Cloud Public trend analysis include government agencies, media institutions, financial industry, real estate industry, education industry, medical industry, tourism industry, and enterprise brand.

 

I. Registration and purchase

Previously, it was made available to the public at the price of 1 RMB per month (UI trial edition) (now upgraded to the UI Public Edition 69 RMB/RMB, the chef tried it at this price. First, open the official link of Alibaba Cloud Shujia, and then click "buy now" on the left of the webpage ", you can register, pay, and use it.

 

II. Use process -- taking Wu Yifan's recent event as an example

After registering and activating the "public trend analysis" service, you can configure the detection topics, keywords, and parameters and analyze the collection content change trend.

(1) set monitoring topics and keywords

When you enter the "public trend analysis" console for the first time, you need to configure the monitoring topics and keywords for the analysis objects. You can use one or more keywords to describe the monitoring topics.

The official Shujia website "quick start" said:

"The system background collects the content of articles that contain the combination of these keywords for summary and analysis. In the future, some statistical functions will focus on monitoring topics. Therefore, we try to create a separate monitoring topic for each analysis object ."

The chef of gossip opened a special topic for the hot Wu Yifan incident and set a keyword combination. First, click the keyword of background management, then click add topic, and then configure the keyword.

In one breath, the chef added a series of hot words such as Wu Yifan, Xiao GNA, Zhuo Wei, and gunner Canada, and selected all the given origin sites including news, forums, Weibo, and WeChat, the topics generated are as follows:

According to official instructions, the key points of keyword configuration are as follows:

"When configuring keywords, we mainly consider the following two factors. (Note: The System Background updates the global keyword collection policy every 10 minutes. Therefore, it takes about 10 minutes to apply the new keyword configuration .)

○ Collection scope: the types of websites on which content needs to be collected. Origin site types can include news, forums, Post bars, Weibo, WeChat, government websites, and video websites.

○ Combination technique: it may take some time to adjust the combination of keywords, whether or not to add synonymous, near-definition, generic leave, or variants of online terms. In addition, do not set a keyword combination that is too broad, such as a single word such as "network" or "security". This will collect too many irrelevant articles, resulting in the rapid depletion of your collection quota. As shown in the following figure, you can add multiple keyword combinations at a time. Each row represents a keyword combination. The number of keyword combinations (that is, the number of rows) varies with the version you ordered. A keyword combination can be composed of one or more words. Words and words are separated by spaces to indicate the relationship between "and: only information containing all words in a row can be collected."

(2) view the collected content

Now you can start to focus on public opinion dynamics. Click "new public opinion" or "all public opinion" in the left-side navigation bar to view the list of articles collected by keywords. The title font in the list is bold to indicate that the document has not been read. Clicking an article will show the details of the article on the right.

The chef randomly opens a text. As shown in the figure below, the emotion has been automatically set as "negative", while the label and emotion (positive, neutral, and negative), risk level, remarks and other information can be customized, at the same time, the chef sets this information to valid.

It is worth noting that marking training is a machine learning process. According to official instructions:

To mark training, you can manually select the article title and mark it as "valid", "invalid", or "read" in batches '. 'Invalid' indicates that these articles are worth attention, and 'invalid' indicates that these articles are interference information. The manual marking process is actually an intelligent classification model in the training background. Generally, after about 100 Mark trainings, the classification model of the system will become more and more accurate, you can use the 'filtering rules' to implement smart filtering. When irrelevant content (such as advertisement, interference information, and spam information) is collected in the future, it can be automatically classified into the 'Recycle Bin '."

The application also provides the article search function. You can click the search button above the article to filter the article from multiple dimensions.

(3) view the result analysis report

Finally, click "homepage" in the left-side navigation bar to view the analysis results of "Today", "7 Days", and "30 Days.

III. Result Display

The chefs checked the results of the incident from October 11 to October 24.

The source site type and keyword cloud analysis are shown in the figure below. The results show that Weibo is the main field of public opinion for this event, which is in line with everyone's potential cognition.

However, there are doubts about the number of public opinions collected. The number of posts posted by searching for "Wu Yifan" on Weibo is far more than 20 thousand articles collected by "public trend analysis" within four days. This shows that the data filtering of the application has its own set of standards, and the sample size obtained by this standard is obviously small. The collection capability of "big data" needs to be observed later.

Result:

In this public sentiment analysis of Wu Yifan's negative event, the average value from 6.21 to 6.24 is 1. The positive public sentiment has a slight upper hand, indicating that a large number of fans or even brains continue to carry out public opinion offensive.

The public's positive sentiment rose sharply on the 1949th, which is consistent with the case of Wu Yifan's reputation infringement.

Taking the Thursday 6.24 as an example, the views on this event were relatively neutral throughout the day, but they reached the emotional peak from to in the morning, the confrontation between the two sides is dominated by supporters (opponents may all be asleep ).

In a word, the chef's automatic emotional judgment on the application's publishing is quite convincing. Although there are some biased articles, most of them can be accurate and qualitative.

The analysis result shows 6.21-6.24 days? -- Top 10 hot events are:

According to the above chart, the subsequent fermentation and trend of Wu Yifan's incident are related to some hot events and hypes-Wu Yifan's reputation infringement case ranks first two in the hot event, the filing on the afternoon of the 1949th pushed the peak of public opinion. This shows that Wu Yifan's public relations practice has played an important role in influencing and leading fans to carry out public opinion counterattack.

IV. Functional comments

Advantages:

1. Excellent emotional analysis, which can accurately show the trend of public opinion.

2. Automated operations and Detection. The interface is simple and easy to use.

Disadvantages:

1. Currently, the application platform is not mature and there are some bugs.

A) after deleting a set topic, the keywords in the topic will not disappear on the public opinion interface, or the data will continue to be loaded on the cloud and cannot be deleted.

B) the analysis results of the home page are not obtained based on a single topic, but the summary analysis results of all keywords and cannot be set. This is unreasonable at 01:10. For example, if you have two topics for detection: "Jingdong 618" and "Wu Yifan", the sentiment analysis and word cloud statistics on the homepage may be a collection of all the moderation texts, currently, differentiated results of different topics cannot be displayed. Therefore, the platform is applicable to an audience with a single detection topic.

2. The volume of captured data is incomplete, and the number of public opinions can easily reach the upper limit. (The chef only detected Wu Yifan's event for four days. The following is a prompt on the console .)

 

3. A number of more practical tools are not yet open to the public. As shown in the following figure, the propagation path analysis function under the open interface is not yet available, so data usage and analysis have limitations.

PS: The chef found that the 1 yuan trial version of the previous few days has been quietly removed from the shelf and replaced by the popular version of 69 yuan/year. According to the person in charge: the trial version of 1 RMB is the activity price of the previous month. It controls the amount of data and can only capture 10 thousand pieces of information every day. Its function is weak, 0.3 million pieces of information can be crawled every day, which greatly increases the analysis capacity of the platform and is truly commercialized. This partially answers the confusion about the chef's experience. Of course, the chef has just experienced the basic edition. Everyone boys or girls is worth having the public edition!

Public trend analysis Address: http://click.aliyun.com/m/5647/