Appen Typeahead Project
This article goes over the guidelines of the project at Appen called Typeahead e2e. This project is somewhat the same as being a search engine rater and rating queries and results, but you are rating queries for Facebook and their website search engine. They may be trying to get this search engine as tuned as Google, except it’s based on their platform. Facebook is considered a family website, and they are trying to make sure that none of the bad stuff gets through. The main purpose of the job you perform is to detect low quality suggestions, and the ratings tasks are mostly based on your own point of view, which is unusual for a project of this type.
What is Typeahead e2e?
Requirements and payment rates
E2E stands for ‘end to end’ but that seems irrelevant at this point. In order to work on this project, you must have a Facebook account. The project has a work volume from 1 to 4 hours per day. This is the usual amount of time on Appen projects. The pay range for this project depends on where you live, but it can range from $6 an hour to $14 an hour. This project is exclusive of Appen and runs in many markets.
Qualification process is straight-forward, before proceeding with actual exam you will be given a chance to do a practice test which consists of 20 tasks and covers all possible rating scenarios. Once you complete the practice set you will receive feedback and recommendations on your ratings. The next step will be actual exam which also consists of 20 tasks. If you pass this exam successfully you will be moved to production.
Typeahead are queries that are given with suggestions when you type something in the search box on Facebook.
The search Typeahead provides suggestions, and it’s your job to see if the suggestions are relevant, or if suggestions are bad. There are a lot of different things that can make them bad. You might recall that Google at one point had a big problem many years back when Google ads were coming up with some violent and hateful content that caused them to lose some advertising with big names. It wasn’t until this happened that Google wanted a real human to look at what ads were being displayed, and what websites were providing as far as search content. The same thing is happening at Facebook and they want to make sure that people are not getting bad results with violent, offensive, hateful, or pornographic content. That is where the human rater comes in, and it’s your job to determine if the results are good or bad, or if they should be flagged.
As mentioned previously, the workflow of this project is to determine the quality of suggestions in the search bar drop down. There are some things that the guidelines go over.
You are presented with a list of categories when doing the search. You make a decision based on what the results are. You would follow these labels and determine what is acceptable and what is offensive/porny (cute new word that Appen or Facebook apparently made up), junky, or acceptable.
It is important to remember that you are rating the quality of the suggestion, and not the quality of the query. Quality doesn’t have anything to do with the query. The query should be looked at to determine if it’s in a foreign language other than the language you are contracted to work in.
Furthermore, a suggestion should be rated 1 if it is offensive, violent or porny, such as “hot girls”, “sexy videos”, etc. The suggestion, at first glance, needs to be obviously offensive or seeking offensive material.
You need to remember that you are not rating the results generated by the Facebook search engine but the suggestion itself. Many common names may generate pornographic results (ex.“mary ann”) but should not be rated as offensive. If it’s junky, then it is because the suggestion is unrecognizable and, most importantly, you think the suggestion is junky/gibberish. Junky entails the suggestion is difficult to understand or is entirely nonsense. If you rate the suggestion as junky, you would then choose a category to support your decision.
A suggestion gets a 3 – Acceptable if it is not junky. This is considered a good suggestion which doesn’t contain foreign language or anything offensive. It is spelled correctly and free of grammatical errors. If a suggestion is rated acceptable, then it must receive an additional rating to indicate whether it is personally recognizable to you. This is a weird task, because you wonder how it can be acceptable if it’s not recognizable. They clear that up by saying ‘The suggestion is UNRECOGNIZABLE based on your own point of view but still makes sense’. They say that you can use Google/Bing to determine if the suggestion is acceptable or junky. But once again, the rating is based on your own personal view.
The job itself doesn’t look very hard, and the work flow should go pretty fast since you only have to choose from 4 different categories, and then choose a label from one of those 4 categories. There are a lot of examples in the guidelines that show you what to do in certain cases in case you are unsure of what to select.
It is interesting to know what this will end up doing to the Facebook pages that rely on certain words that are brought up. Some searches may end up disappearing from the suggestions at all.
As always, if you have any questions about this project, or you are working on this project, leave us a comment and tell us how it’s going.