Analysis of Weibo Search Strategy

The shortcomings of Weibo search in different scenarios are analyzed based on two aspects: user scenarios and the satisfaction of display results. The program intelligently provides personalized solutions to meet users' needs for efficient information acquisition. This article is an analysis of Weibo search strategies, enjoy~

1. Research Objectives

Study the search effect of Sina Weibo in different demand scenarios and propose optimization solutions to meet users' needs for efficient information acquisition.

2. Research and Analysis

Referring to the thinking method of web search strategy, the shortcomings of Weibo search will be analyzed from the perspective of demand clarity, query characteristics and requirements for results:

Figure 1 Survey Analysis

1. Clear requirements

Referring to the thinking method of web search strategy, the shortcomings of Weibo search will be analyzed from the perspective of demand clarity, query characteristics and requirements for results:

(1) The query structure is simple and clear

Scenario 1: Search for "My Neighbor Totoro"

Search term: My Neighbor Totoro movie

Expected results: Discussion about the new movie My Neighbor Totoro

Actual results show:

Search results analysis:

The search results meet the basic needs of users. The first one is an introduction to the movie "My Neighbor Totoro". The second to fourth ones are reviews from well-known bloggers. However, the information below only contains the word "My Neighbor Totoro", such as the complaints and daily statements of celebrities and bloggers, and the bloggers' film and television recommendations. There is no other valid information about "My Neighbor Totoro".

From the query, we can see that Weibo has segmented the search terms (some do not contain the word "movie"), but has not performed corresponding semantic analysis. In the first four pieces of information, the popularity and relevance are high, which can meet the needs of most users. However, in the following information, as long as there are relevant keywords in the query segmentation that match the latest news, the system will rank the information based on popularity, which will lead to a decrease in relevance and the push results will not meet user expectations.

Scenario 2: Search for "actor Song Zhongji"

Search term: Actor Song Joong-ki

Expected results: The latest developments of Song Joong-ki

Actual results show:

Search result analysis: The search results basically meet the needs of users. The first one is a hot topic on the official Weibo account, which is the latest news about Song Joong-ki and meets the needs. Looking further down, most of the information still meets the needs, but there is still a problem with word segmentation. For example, the blogger's daily statements contain the word "Song Zhongji", but there is no other valid information. This is also a problem of query word segmentation matching. This is problem one.

In addition, there is a large amount of duplicate information under each topic that has not been systematically screened, which increases the cost for users to obtain differentiated information. For example, the news that the filming of the movie "Bogota" starring Song Joong-ki was suspended was pushed frequently and continuously by different bloggers. This is the second problem.

(2) Colloquial query

Scenario 3: Search for "What medicine can cure a cold quickly?"

Search term: What medicine can cure a cold quickly?

Expected results: Recommended information on cold medicines

Actual results show:

Search result analysis: The search results are basically satisfactory.

The first one recommends a popular article that matches the query very well. The second one recommends a popular Q&A video whose title matches the query very well. The third one recommends an industry official microblog that directly provides a solution that matches the query. It can be said that the first three items meet the needs very well.

By the fourth, fifth, and even second articles, a large number of bloggers recorded their daily lives, and the words such as cold and taking medicine were included in the records. This shows that the system only did a simple keyword match and did not understand that the user’s real need is to find medicine to treat colds. This is the first problem.

What’s worse, after the tenth item, the information “How to treat a chicken’s cold? What’s the best medicine for a chicken’s cold?” appeared. This shows that the system did not perform semantic reasoning, that is, the user did not add a subject such as “person” in front of the search, so it should be assumed that a person has a cold. If it is an animal, the user will definitely add cats and dogs when searching. No semantic analysis was performed. This is the second problem.

In addition, from the query perspective, the word segmentation is "cold+eat+medicine+what+good+fast", the query matching degree is very high, and there is no obvious loss or reorganization of the word segmentation. However, the system does not do in-depth mining of the search needs behind the query. "Cold" is the symptom, and "medicine" is the method. "What" is added in front of it, which obviously means it is looking for a solution. Of course, it is also a problem of semantic analysis.

Scenario 4: Search for “How much does it cost to straighten teeth”

Search term: How much does it cost to straighten teeth

Expected results: Recommended price information for orthodontic treatment

Actual results show:

Search results analysis:

Searching for "how much does it cost to straighten teeth" only resulted in three messages being pushed out on Weibo:

• The first one is an advertisement, and it is an advertisement that is completely unrelated to the "query". It does not meet the needs, and the user needs to turn the page to continue to find relevant information, which increases the cost of finding information answers. This is the first problem;
• The second one is an article that matches the query very well, but the content of the article is limited to Luohu, Shenzhen, and is not comprehensive enough;
• The third one is the blogger’s request for help, but he also received no answer and his request was not met.

This shows that there are very few information resources about “price of orthodontic treatment” on Weibo, or the relevance of the system design is too low to make it recommended.

(3) Complex query expressions

Scene 5: Search "In which episode did Shinichi confess his love to Xiaolan?"

Search term: In which episode did Shinichi confess his love to Xiaolan?

Expected result: I want to know which episode "Shinichi confesses his love to Xiaolan" is in

Actual results show:

Search result analysis: It can basically meet the needs, but there are two problems:

1. The first two pushes are still the blogger's daily content containing relevant information with the same keywords as the query. Netizens give answers, but users cannot immediately see the results through the summary. They need to click again to continue searching for answers from top to bottom. Although this process meets the needs, it still increases the cost of obtaining effective information.
2. The next page is another advertisement, so users still need to scroll down to continue looking for relevant information, which increases the cost of finding high-quality answers.

Scene 6: Search for the line "I'm such a fool, I can't do anything except liking you" from which movie

Search term: I am such a fool. I can't do anything except liking you. Which movie is this line from?

Expected result: I want to know the source of this line

Actual results show:

Search result analysis: The first two items can meet the needs, but they are basically indirect answers to the questions. The third, fourth and even tenth items are the blogger’s inner thoughts, which are just a simple match to the query. Moreover, from the results we can see that the query is segmented into "I'm so stupid, I can't do anything except liking you" + "is" + "which" + "movie" + "in" + "of" + "lines". Obviously, there is a clear loss phenomenon.

2. Clear requirements and special requirements for results

Scenario 7: Search for "Latest Housing Prices in Beijing"

Search term: Beijing latest housing prices

Expected result: Want to know the latest housing prices in Beijing

Actual results show:

Search results analysis:

The demand is met well, namely Beijing, latest, and housing prices. The pushed results are not only related to Beijing's housing prices, but also meet the special demand of "latest". The author believes that this is related to the fact that the query is a hot topic and that there are also official information resources in the system.

Scenario 8: Search for "Latest Housing Prices in Beijing"

Search term: Huawei's latest mobile phone

Expected result: Want to know the latest information about Huawei's mobile phone

Actual results show:

Search results analysis:

The demand is not met well. The first push is hot news about Huawei, which does not match the user's demand - the latest mobile phone. This shows that the system's recommendation relies too much on hot spots, which is the first problem. The second push is the function and price of Huawei's latest mobile phone, which meets the demand. The third push is a piece of information with semantic understanding. The query is segmented into "Huawei + latest + model + mobile phone". Obviously, there is a reorganization after segmentation, but the result basically meets the demand. The fourth push is an irrelevant advertisement. The fifth push is a popular article with a very high degree of match with the query, which meets the demand, but you need to click in to get the answer.

In general, search results depend on popularity and update time.

3. Unclear requirements

Scene 9: Search for "Qiu Ming"

Search term: Qiu Ming

Expected results: Learn about He Youying's painting "Autumn Darkness"

Actual results show:

Search results analysis:

Basically does not meet the needs. The system doesn’t know what “Autumn Darkness” is? The best result is finding information matching the keyword "Qiu Ming", but it is not the result the author expected.

Scene 10: Searching for "Three Body"

Search term: three body

Expected results: Learn about the book "The Three-Body Problem"

Actual results display

Search result analysis: basically does not meet the needs. The first three push notifications only contained the word "Three-Body", which had nothing to do with the content of the book "Three-Body"; the fourth push notification even contained an advertisement that had nothing to do with "Three-Body"; the fifth push notification finally contained information about the book "Three-Body" and its author; however, the sixth push notification did not match the expected result. Therefore, if you want to obtain accurate information, you need to expand the search terms.

From the research and analysis of the above 10 scenarios, we can see that the shortcomings of Weibo search are concentrated in the following aspects:

• Recommended order: The popularity and latest time weights of Weibo are too high, and the matching degree of the query is ignored.
• Information duplication: Too much repeated information increases the cost for users to obtain differentiated information within a limited time.
• Semantic understanding: It mainly relies on the matching degree between the query and the text information of the information resource, but cannot understand the real needs behind the query.
• Advertisements are pinned to the top of the page: Weibo’s excessive promotion of advertisements distracts users from efficiently obtaining information.
• The processing method for popular queries and unpopular queries is simple: they are basically pushed based on the principles of popularity, latestness, and query matching, resulting in a large amount of invalid information in the push results.
• Semantic conversion: There is a big difference between the spoken query and the written query, such as: "How much does it cost to straighten teeth" and "the price of straightening teeth".
• Display format: Some answers are not clear at a glance. Some popular articles and Q&A topics include query search answers, but the search results only have titles and no abstracts. The answers cannot be found all at once, and users must click in to continue selecting. In addition, there are no markings for related keywords, such as red bleaching.

3. Optimization plan

Based on the above survey results, we summarize the shortcomings of Weibo search in different scenarios, break down the problems based on user scenarios and the satisfaction of display results, and use program intelligence to provide personalized solutions to meet users' needs for efficient information acquisition.