Wouri project review
There’s another project out by Appen called Wouri. We’ll go over the project a little and briefly touch on the main points, what type of job it is and what requirements there are. We will also touch on the grading process and give a few examples.
The goal of this project is to understand the user’s intention as represented by the query, and then determine how good the result is and if it satisfies the user’s needs. Users are using Safari and Spotlight to search for different information needs. Some examples include wanting to find out knowledge or information on something or somebody, wanting to know if there is a certain store nearby, the latest news related to a topic, or since they are MAC users, getting an application on iTunes. This project is a bit like the average search engine evaluator, but it does have a different grading system. However, it uses some of the same ideas from those programs on search engine results.
A desktop or laptop computer is recommended over a mobile phone and all graders should be checking search results with Safari. Safari can be used on Windows operating systems, but it should be mentioned that there is no support from Apple when using Safari on MS Windows. You might be better off with a MAC computer, but it isn’t required.
To participate in the project, you have to pass an initial assessment and achieve a minimum score of 80%. If you don’t pass the first time, you’ll get another attempt. After you pass the initial assessment you will be given a final assessment in which you must also pass with a minimum score of 80%.
Wouri Project rating Logic
When looking at a user query, there are three steps that you will take in the grading process.
- The first thing is understanding the query. You should make sure it’s not too vague, inappropriate, or in a foreign language.
- The second thing is understanding the result. Determine if it is GEO or Time Sensitive. There is also the option to choose both or choose neither GEO or Time Sensitive.
- Once you understand the query and the result, you will then decide on a relevance rating for the result based on the query intent.
It is important to note that Appen expects local graders to judge the search results of the same locale. What that means is you must in the same location as the project requires. In this case, you may need to configure your computer and your search engine to be in the language of the project that you are working on.
A web relevant search is one that satisfies the user’s information need. Of course, a search result can be more or less relevant based on how well or how completely it satisfies the users’ needs. It is important, and we cannot stress this enough, that you have the guidelines open and you have thoroughly read them.
In step number one, you need to determine what it is that the user is searching for and decide if it’s valid. You should read the query, and review the third party results and think about the possible user intent that it represents. This is called ‘Understanding the Query.
For example, the query ‘Samsung’ could have many different user intents such as shopping for Samsung products, getting Samsung technical support, or the user may be looking for the Samsung homepage. However, if the user simply types in Samsung, then the query can be too vague.
If you have no idea what the user might be looking for, then the next step would be to go to a popular search engine and study the term. Some common terms could have some special meaning. However, if it doesn’t have some special meaning, or it has too many special meanings with each one having a small portion of intent, then you would select Query is Too Vague.
You might also run into a situation where a query is inappropriate and it encourages illegal behavior, includes profane words, using derogatory language, racial slurs or hate speech. It also includes graphic violence or the query might like to retrieve grisly content. You are not expected to do anything about these types of queries except to mark it as inappropriate.
When in doubt about any type of query, you should check results from other search engines. It’s absolutely crucial to understand what the user is searching for, and you may need to check results from other search engines. The good thing about this project is that top results from other search engines are provided for you. There is a small link that says ‘Web Search Link’ that will automatically do this for you and show you the top results in order to better help you understand the query.
It is VITAL to point out that opening the webpages is a MUST when you are working on this project. If the results match the first result of the other search engine, then the result is likely an excellent match. Make sure to check the results that are returned by the Web Search Link. Each result has a clickable URL, a small description, and a photo where applicable.
We touched a little on GEO and Time Sensitive in the beginning, and since you might not understand what that is, we will go over it to help you understand it better.
If a query is about a store or some location that exists in other locations aside from the searcher, let’s say a Starbucks, or a query like ‘movie theatres near me’ then you know its geo sensitive meaning that it matters where and how close the result to the searcher is. If the searcher is from California and looking for a Starbucks and the first result is the homepage or another location, then it doesn’t satisfy the user’s needs.
Anything that deals with a query that involves some kind of time frame, such as ‘Buy Superbowl Tickets’, age of Harrison Ford, or checking on the weather are results that should be marked as ‘Time Sensitive’.
It is important to understand that a query can be both Geo sensitive and Time sensitive, and a good example is a query such as ‘tickets to the local basketball game’, local movie showtimes, etc.
Once you understand the query and you judge the results, then you will assign a relevant grade for each result. There are 9-grade levels:
● Query In Foreign Language
● Result Page In Foreign Language
There are many examples in the guidelines that go over difficult queries in order to help you evaluate what results would satisfy the informational need for the searcher. It is a must that you go over the guidelines and read them and understand them so you know how to grade something.
The guidelines go over when you should choose an excellent result, and the biggest point to keep in mind here is to believe that the result will be everything that the user will need, and they will not have to continue searching. If you believe that the result isn’t good enough, and you would also do an additional search, then you might choose ‘Good’ or ‘Fair’ as the grade.
A ‘Good Result is one that the user wants to see above the fold on the device screen. It is typically the first 4-5 results. However, it is not the best possible result, which would be excellent, but it still provides good information. The guidelines have many examples of what should be chosen as a ‘Good Result’.
When choosing a ‘Fair Result’ for the query, the result would only be somewhat relevant, but not be of interest to all users. An example of this would be a query for a celebrity name, but the result is an obscure person that has the same name.
Choosing ‘Bad Result’ would pertain to a result that does not satisfy user intent and is something that should not be shown to users at all. An example of this would be something that we previously mentioned before if a user was looking for a Starbucks Coffee and lives in San Francisco, but the map result is for a store in San Diego. There are other examples in the guidelines.
It should be straight-forward of when you should choose ‘Foreign Language’. If the given query is in a different language from the specified one, or the target language that you are assigned to work, then you would mark this as “Query in Foreign Language”. If you are not sure if the query is in a foreign language, then just search that query to understand the word or words.
Sometimes in the given search results for a query, the URL will be inaccessible, or website link will not be available or a dead link. If this happens then you would grade it ‘Unavailable’ This also includes landing pages that take longer than 10 seconds to load, or you may see that the domain is up for sale by being sent to a redirection page. This grade also pertains to when video content is unavailable or blocked.
There are some other specific result types that the guidelines go over including ‘Ambiguous Queries’ or when the interpretation is much more popular than others. For example, the query is ‘Golden Retriever’ but the result is a song titled the same. This should be graded as fair because the dominant interpretation would be for the dog breed itself and not a song title. There are some instanced when multiple interpretations can be as equally popular though. In order to assign something to the excellent category, it’s a good idea to take into consideration what geo-location the seacher is looking from. Example:
Query is “UM football”, and the result is the official website for the University of Miami football team. If user’s location is in San Francisco, grade the result as Good, since UM could refer to either U of Miami or U of
Michigan and we don’t know which is best. But if the user location is in Miami, grade as Excellent.
As always, if you have any questions about the program, please send an email or reply in the comments. If you are working on this program, then we would love to hear about your experience in the comments.