Homework 2: In-Memory Search Engine

Part A: File Parser

You're going to write a module that reads the contents of a text file into memory and then parses the text file to look for words within it. As it finds words, it will build up a HashTable that contains one record for each word. Each record will contain a lowercase copy of the word, and also a sorted linked list. Each record of the linked list contains an offset within the file that the word appeared in (the first character in the file has offset zero). However, our word parser won't be very smart. It will treat as a word any non-zero sequence of alphabetic characters separated by non-alphabetic characters.

So, graphically, what your module will do is take a text file that contains something like this, where '\n' represents the newline control character that appears at the end of each line in the input file:

My goodness!  I love the course CSE333.\n
I'll recommend this course to my friends.\n

and produces a data structure that looks like this:

Specifically, note a few things:

• Each key in the hash table is the result of calling the HashTable module's FNVHash64() function, passing the string as the first argument, and the strlen(string) as the second argument.
• Each element in the hash table is a pointer to a heap-allocated structure that contains two fields; a string and a linked list. Note the string is lower-cased, and that our parser is not very smart: because it treats any sequence of alphabetic characters surrounded by non-alphabetic characters as words, the word I'll will be misparsed as the two words i and ll.
• Each element in the linked list is an integer representing the position in the text file at which the word starts; this is both its byte offset and, since we are only handling ASCII files, the number of characters from the start of the file (each ASCII character is exactly 1 byte). So, the word "my" starts at offset 0 in the text file, the word "i" appears twice, once at offset 14 and once at offset 40, and the word "course" appears twice, once at offset 25 and once at offset 62.
• Each list is sorted in ascending order.

Part A Instructions

1. Your job in Part A is to finish the implementation of FileParser.c. Start by reading through FileParser.h and make sure you understand the semantics of the functions. Also, look at the WordPositions structure defined in FileParser.h and compare it to the figure above. The function ParseIntoWordPositionsTable() builds a HashTable that looks like the figure, and each value in the HashTable contains a heap-allocated WordPositions structure.
2. Similar to HW1, look through FileParser.c to get a sense of its layout, and look for all occurrences of STEP X (e.g., STEP 1, STEP 2, ...) for where you need to add code. Be sure to read the full file before adding any code, so you can see the full structure of what we want you to do. Once you're finished adding code, run the test suite and you should see some tests start to succeed!

Part B: File Crawler and Indexer

At a high-level, a search engine has three major components: a crawler, an indexer, and a query processor. A crawler explores the world, looking for documents to index. The indexer takes a set of documents found by the crawler, and produces something called an inverted index out of them. A query processor asks a user for a query, and processes it using the inverted index to figure out a list of documents that match the query.

File System Crawler

Your file system crawler will be provided with the name of a directory in which it should start crawling. Its job is to look through the directory for documents (text files) and to hand them to the indexer, and to look for subdirectories; it recursively descends into each subdirectory to look for more documents and sub-sub-directories. For each document it encounters, it will assign the document a unique "document ID", or "docID". A docID is just a 64-bit unsigned integer.

Your crawler itself will build two hash tables in memory, adding to them each time it discovers a new text file. The two hash tables map from docID to document filename, and from document filename to docID:

For each document the crawler finds, it will make use of your Part A code to produce a word hashtable using ParseIntoWordPositionsTable().

Indexer

This is the heart of the search engine. The job of the indexer is to take each word hashtable produced by ParseIntoWordPositionsTable(), and fold its contents in to an inverted index. An inverted index is just a hash table that maps from a word to a "posting list," and a posting list is just a list of places that word has been found.

Specifically, the indexer should produce an in-memory hash table that looks roughly like this:

Walking through it, the inverted index is a hash table that maps from a (hash of a) word to a structure. The structure (shown in green) contains the word as a string, and also a HashTable. The HashTable maps from the docID (not the hash of docID) to a LinkedList. The LinkedList contains each position that word appeared in that docID.

So, based on the figure, you can see that the word "course" appeared in a single document with docID 3, at byte offsets 25 and 62 from the start of file. Similarly, the word "love" appears in three documents: docID 1 at positions 7 and 92, docID 3 at position 16, and docID 4 at positions 18, 21, and 55.

Part B Instructions

The bulk of the work in this homework is in this step. We'll tackle it in parts.

1. Take a look at DocTable.h; this is the public interface to the module that builds up the docID-to-docname HashTable and the docname-to-docID HashTable. Make sure you understand the semantics of everything in that header file; note how we can now implement procedural-style class composition! We create a single DocTable structure that contains both of these tables, so when you implement DocTable_Allocate(), you'll end up malloc'ing a structure that contains two HashTables, and you'll allocate each of those HashTables.
2. Take a look inside DocTable.c; this is our partially completed implementation. Be sure to read the full file. Your job, as always, is to look for the "STEP X" comments and finish our implementation. Once you've finished the implementation, re-compile and re-run the test_suite executable to see if you pass our tests. If not, go back and fix some bugs!
3. Now, take a look inside CrawlFileTree.h; this is the public interface to our file crawler module. Make sure you understand the semantics of everything in that header file. Next, read through the full CrawlFileTree.c and then complete our implementation. Once you're ready, re-compile and re-run the test_suite executable to see if you pass more tests. If not, go back and fix some bugs!
4. Finally, take a look inside MemIndex.h. This is the public interface to the module that builds up the in-memory inverted index. Make sure you understand the semantics of everything in that header file and note how we implement procedural-style inheritance! Next, read the full MemIndex.c, and then complete our implementation. (This is the most involved part of the assignment.) Once you're ready, re-compile and re-run the test_suite executable to see if you pass our tests. If not, go back and fix some bugs!
5. Once you've passed all of the tests, re-run the test suites under valgrind and make sure you don't have any memory leaks or other memory issues.

Part C: Query Processor

Now that you have a working inverted index, you're ready to build your search engine. The job of a search engine is to receive queries from a user, and return a list of matching documents in some rank order.

For us, a query is just a string of words, such as:

course friends my

The goal of our search engine is to find all of the documents that contain all of the words. So, if a document has the words "course" and "my" but not the word "friends," it shouldn't match.

To execute a query, first the query processor must split the query up into a list of words (the strtok_r() function is useful for this). Next, it looks up in the inverted index the list of documents that match the first word. This is our initial matching list.

Next, for each additional word in the query, the query processor uses the inverted index to get access to the HashTable that contains the set of matching documents. For each document in the matching list, the query processor tests to see if that document is also in the HashTable. If so, it keeps the document in the matching list, and if not, it removes the document from the matching list.

Once the processor has processed all of the words, it's done. Now, it just has to rank the results, sort the results by rank, and return the sorted result list to the user.

For us, our ranking function is very simple: given a document that matches against a query, we sum up the number of times each query word appears in the document, and that's our rank. So, if the user provides the query "foo" and that words appears on a document 5 times, the rank for that document given that query is 5. If the user provides a multi-word query, our sum includes the number of times each of the words in the query appears. So, if the user provides the query "foo bar", the word "foo" appears in the document 5 times, and the word "bar" appears 10 times, the rank of the document for that query is 15. The bigger the sum, the higher the rank, and therefore the "better" the search result.

Part C Instructions

We have provided a mostly empty skeleton file searchshell.c. It is up to you to complete it by implementing a program that uses the Linux console to interact with the user. When you run the query processor (called searchshell : you can try a working searchshell in the solution_binaries/ directory), it takes from a command-line argument the name of a directory to crawl. After crawling the directory and building the inverted index, it enters a query processing loop, asking the user to type in a search string and printing the results. If the user signals end-of-file when the program asks for a search string (i.e., control-D from the Linux terminal keyboard), the program should clean up any allocated resources – particularly memory – and shut down.

When you are ready, try running ./searchshell to interact with your program and see if your output matches the transcript from a search session with our solution. Alternatively, compare your searchshell to the searchshell we provided in the solution_binaries/ directory. Note that our ranking function does not specify an ordering for two documents with the same rank. Documents that have the same rank may be listed in any order, and that might be different from the ordering in the sample transcript or produced by the solution_binaries version of searchshell.

Bonus Tasks

If you do work on either of the bonus tasks, you must also include a hw2-bonus tag in your repository. While grading, we will use whichever commit has that tag to examine the bonus, so it may be the same or a different commit from the one that has hw2-final. If you do not have a hw2-bonus tag in your repository, we will assume you did not choose to submit anything for the bonus.

1. When searching text documents for a given query, some words in the query do not add any value to the quality of the search results. These words are known as stop words and include common words such as the, is, and, at, which, there, on, and so on. Some Web search engines filter these stop words, excluding them from both their indexes and ignoring them in queries; this results in both better search quality and significantly more efficient indexes and query execution.

   Implement a stop word filter. The search shell should accept a second, optional argument "-s" that will act as a flag for turning the filter on. When the flag is not specified, your program should not filter any stop words. It is up to you to decide how you will implement the stop word filter (and where you'll get your list of stop words), but be sure to create a README.md file and explain in that file what changes you had to make and how your filter works. Stop words that have apostrophes in them should be handled the same way that you've handled the words in the documents.

2. You probably noticed that we went to a lot of trouble to have you include word position information in our inverted index posting lists, but we didn't make use of it. In this bonus task, you will! In addition to letting you search for words, modern search engines also let you search for phrases. For example, I could specify the following query:

   alice "cool fountains" flowers

   This query would match all documents that contain all of the following:

   • the word alice,
   • the phrase cool fountains (i.e., the words cool and fountains right next to each other, in that order),
   • the word flowers.

   Using the positions information in the inverted index postings list, implement support for phrase search. You'll have to also modify the query processor to support phrase syntax; phrases are specified by using quotation marks. Be sure to create a README.md file that describes what changes you had to make to get phrasing to work.