Requirement Analysis Design: A case of requirement analysis in box calculation

Article Description: Overview of requirements analysis in box calculation.

The goal of the box calculation is to provide the user with a one-stop search service based on the search box. For example, when users type in the search box "Non-interference", the system can make it clear that query has to seek TV program video, film video, look up film review, participate in discussions and view the relevant news needs, and then assign these different requirements to the best content resources or applications for processing, Ultimately, the results of satisfying these needs are presented to the user in a precise and efficient manner. Figure 1 shows the "box" to query "Non-interference" to meet the situation. From the analysis above, it's not hard to see that the first step in box computing is to identify what needs to be in a query, which is exactly what the requirements analysis is going to do.

Figure 1 "Non-sincere not disturbing" box calculation results

Requirements analysis is the entrance to the box calculation, because only the need to analyze the query, in order to better show the corresponding results to meet the needs of users. Therefore, the more accurate analysis of query requirements, the coverage of query, the user satisfaction is greater. Requirement analysis is one of the most important and difficult places in box computing.

The easiest way to identify query needs is to check the thesaurus. For example, all the movies, TV series are collected in advance, put into the thesaurus, as long as the user input the word in the glossary, you can identify a query whether there is a need for video. The advantages of this approach are fast, but the disadvantages are obvious: First, this method can only divide query into 2 dimensions, either a video requirement or no video demand. But a lot of video's name, has many meanings, in the video above just one of the faint meaning, if the direct out of the video demand, will greatly harm the user. For example, there is a advertising film named "Baidu", this advertising film may be more than 60% users are unfamiliar, if more than 60% of the user input "Baidu", directly to give a "Baidu click" Video viewing results, then more than 60% of users will feel very confused. Second, this approach does not meet the needs of most users. Users in the search for video, the presentation method is a variety of, such as: Soldiers Blitz complete, soldiers surprise HD, soldier Blitz complete online watch, etc., these 3 query have a strong search for video needs, but by way of look-up table, but can not recognize the demand. Again, this method is not satisfied with the video demand outside the movie and TV play. There are often more than movies, TV dramas, popular or news-time-sensitive things that have a strong demand for video. For example: Xidan girls, Zhongguancun Boys, Diaoyu Islands collision ship, Korean artillery shelling, and so on, most users often have to see the relevant video needs. Finally, this approach lacks predictability and identifies a limited number of query numbers. That is, for the term in the glossary, can identify the video needs, but the glossary outside the term, can not identify the video demand. Therefore, if the vocabulary size is n, you can only identify n query with video requirements.

From the analysis above, we can conclude that a good demand analysis method, at least to meet the following requirements: 1 has a high accuracy and recall rate, that is, in the identification of more than 90% query requirements, but also to ensure that the need to identify, more than 95% are correct 2) has a good predictive ability, that is, can accurately predict the demand of unknown query, 3 has the ability to identify requirements and analytic fields, that is, in the identification of query requirements, but also to effectively resolve the required information.

Here are a few typical examples. Through these examples, you can have a general understanding of the tasks to be accomplished in the requirements analysis.

In Figure 2, the requirement analysis needs to be done to parse out the demand for query with Exchange rate conversions, and to parse out the amount of 56.7, and the source currency is USD and the target currency is RMB.

Figure 2 showing the exchange rate results directly

In Figure 3, the requirements analysis needs to identify that query has the need to find a train timetable, and to resolve the starting point for Beijing, the terminus of Shanghai. Then directly for users to show the train timetable from Beijing to Shanghai, greatly facilitate the user to obtain relevant information.

Figure 3 Direct information on train trips

Figure 4, Figure 5 shows how the requirements analysis can identify query with the need to find uncommon words and unit conversions, and directly give the answer to the process. This process is not visible to the user, but for the system, it requires a series of complex identification, analysis and presentation.

Figure 4 Meet the requirements of uncommon Word query

Figure 5 Meeting Unit conversion requirements

Figure 6 shows a requirement analysis that identifies a user's query with the need to seek job information, resolves the company name, Posts 2 fields, and then retrieves a sample of the jobs that are displayed to the user directly from the relevant data source.

Figure 6 Meeting user search job requirements

As you can see from the analysis and examples above, the goal of query requirements analysis is to identify the intended or interested areas of query and extract the required fields from it. It is a very challenging job to accomplish this task. First of all, the diversity of user query presentation has brought a lot of difficulties to demand analysis. For example, there are hundreds of different ways to express a need for weather. Second, similar query may have different requirements. For example: "From Beijing to Shanghai car distance" and "from Beijing to Shanghai ticket", the former has a strong self-driving demand, and the latter has a strong demand for train time. And the same star, Faye Wong, Jacky Cheung, have a strong video, pictures and other needs, and Lee Kai-fu, Zhang and other technology stars, from the needs of users, the most strong demand is to find encyclopedia information, not video, pictures. Finally, user-entered query is often expressed in natural language, sometimes including abbreviations and spelling errors, such as: 5 singapore dollar =? Dollar, 5 personal coin equals how many dollars. These objective problems have increased the difficulty of demand analysis.

To better meet the diverse needs of users, strong demand analysis is necessary. Baidu box calculation needs analysis, is through the semantic analysis of query, combined with user behavior analysis and massive computing technology to achieve. Through various means to evaluate, has already achieved very good effect. For example, by analyzing the query tens of billions of times a day, the user retrieves using machine learning methods, we can automatically learn from the user's various idiomatic or obscure expressions, combined with semantic analysis, information extraction and other technologies, can accurately and efficiently identify the needs of query and resolve the required information, For the majority of netizens to provide the most convenient search service and search experience.