I Running the web service on Mac OSX

1. Clone trolldetector from https://github.com/tikabom/trolldetector
2. Execute TrollDetector.exe @ trolldetector/TrollDetector/bin/Debug
3. If step 2 fails download Mono from http://www.mono-project.com/docs/getting-started/install/mac/
4. In your console run this command: mono "Path to TrollDetector.exe"

II Testing the web service

1. To obtain a sample text and list of words to be excluded, run the service and make a GET request to the URL localhost:8000/TrollDetector/
2. To test word count, run the service and make a POST request to the URL localhost:8000/TrollDetector/ with a JSON object in the Request body that looks like:
   {
   Text: "some text",
   Exclude: ["some"],
   WordCount: {
   "text": 1
   }
   }
   Assumption: The service expects the client to send its request in JSON containing three properties: Text (text used for the word count), Exclude (list of words that should not be counted), WordCount (every word that is counted and its corresponding count).

III To return sample text and list of words to be excluded:
The service has access to files that contain text. When a client requests for some text from the service to perform word count on, the service randomly picks a file. The text in the file is parsed to extract a list of unique words. If the text contains only one unique word, the text is returned to the client with an empty list for the exclusion words. If there are more unique words, the number of words N to be added to the exclusion list is chosen at random based on the number of unique words L. N words are chosen at random amongst the L unique words and returned to the client.

IV To test count of words returned by the client:
The client sends the text and list of exclusion words it received, along with the word count it calculated. The service tests the data it receives in the following order:

1. Checks the text against all the files it has access to. If the text is not from any of these files, it returns 400 Bad Request to the client.
2. Based on the exclusion list, the service counts words. If there is only one unique word in the text and the exclusion list submitted by the client is not empty, the service returns 400 Bad Request to the client.
3. If the number of words the service has counted is not the same as the number of words submitted by the client, the service returns 400 Bad Request to the client.
4. If a word counted by the client is also in the exclusion list, the service returns 400 Bad Request to the client.
5. If the count returned by the client is not the same as what the service computed, the service returns 400 Bad Request to the client.

V Unit Testing: All unit tests for the service have been written in the project UnitTestTrollDetector.

VI Ideas:
This service is currently stateless. In a real-time scenario, state can be maintained in a database. Assuming there are several text files available to send to requesting clients, each time a text is sent, it is recorded in a table with its exclusion list and response timestamp. An appropriate timeout is defined for all clients to return their word count response. A scheduled job at the server checks these timestamps and deactivates records for which the timeout has elapsed. When a new client sends a request for text, it is sent text that is not currently active in the database. When a client responds back with its word count, the text and exclusion list it sends is verified against the active records. If the service doesn't find a matching active text and exclusion list, it sends the client a 400 Bad Request response. Else, it follows all the steps mentioned in the section IV.

VII Data structures:
HashSets are used to create lists of unique words and the exclusion list. While creating the exclusion list, there is a possibility for duplicates when a word is randomly being chosen from a unique set of words. This is handled by the HashSet. Dictionaries are used to save word counts. Each key-value pair in the dictionary represents a unique word that is not in the exclusion list and its corresponding count in the text.