School of Computing

Information Retrieval

NUS SoC, 2013/2014, Semester II Video Conferencing Room (COM1 02 VCRm) / Fridays 11:00-13:00

Last updated: Wednesday, March 12, 2014 08:32:43 AM SGT - Revised ddd.qqq scheme and gave explicit output format.

Homework #3 » Vector Space Model

In Homework 3, you will be implementing ranked retrieval described in Lectures 7 and 8.

Commonalities with Homework #2

The indexing and query commands will use an identical input format to Homework 2, so that you need not modify any of your code to deal with command line processing. To recap:

Indexing: $ python -i directory-of-documents -d dictionary-file -p postings-file

Searching: $ python -d dictionary-file -p postings-file -q file-of-queries -o output-file-of-results

We will also again use the Reuters training data set provided by NLTK. Depending on where you specified the directory for NLTK data when you first installed the NLTK data (recall that the installation is triggered by, this data set is located under a path like: .../nltk_data/corpora/reuters/training/.

As in Homework 2, your search process must not keep the postings file in memory in answering queries, but instead use a file cursor to randomly access parts of the postings file. However, there is no such restriction for the indexer (you are not asked to implement BSBI or SPIMI). Also, you should continue to employ Porter stemming in your submission, both for document indexing as well as query processing.

Vector Space Model

To implement the VSM, you may choose to implement (you can can do it differently) your dictionary and postings lists in the following format. The only difference between this format and that in Figure 1.4 in the textbook, is that you encode term frequencies in the postings for the purpose of computing tf×idf. The tuple in each posting represents (doc ID, term freq).

term doc freq (df)
postings lists
ambitious 5 (1, 5)→ (7,2) → (21, 7) → ...
... ... ...

In the searching step, you will need to rank documents by cosine similarity based on tf×idf. Bug fixed! In terms of SMART notation of ddd.qqq, you will need to implement the ltc.lnc ranking scheme (i.e., log tf and idf with cosine normalization for documents, and log tf, cosine normalization but no idf for queries. Compute cosine similarity between the query and each document, with the weights follow the tf×idf calculation, where term freq = 1 + log(tf) and inverse document frequency idf = log(N/df) (for queries). That is,

tf-idf = (1 + log(tf)) * log(N/df).

It's suggested that you use log base 10 for your logarithm calculations (i.e., math.log(x, 10), but be careful of cases like math.log(0, 10)). The queries we provide are now free text queries, i.e., you don't need to use query operators like AND, OR, NOT and parentheses, and there will be no phrasal queries. These free text queries are similar to those you type in a web search engine's search bar.

Your searcher should output a list of up to 10 most relevant (less if there are fewer than ten documents that have matching stems to the query) docIDs in response to the query. These documents need to be ordered by relevance, with the first document being most relevant. For those with marked with the same relevance, further sort them by the increasing order of the docIDs. Updated Output these documents as a space delimited list as in Homework #2. Be sure not to include extra whitespace before and after any document IDs. For example, if document IDs 2, 1024 and 512 are only 3 relevant documents, in that order your searcher should output these document IDs one one line for the corresponding query:

2 1024 512

What to turn in?

You are required to submit,, dictionary.txt, and postings.txt. Please do not include the Reuters data.

Essay questions

You are also asked to answer the following essay questions. These are to test your understanding of the lecture materials. Note that these are open-ended questions and do not have gold standard answers. A paragraph or two are usually sufficient for each question. You may receive a small amount of extra credit if you can support your answers with experimental results.

  1. In this assignment, we didn't ask you to support phrasal queries, which is a feature that is typically supported in web search engines. Describe how you would support phrasal search in conjunction with the VSM model. A sketch of the algorithm is sufficient. (For those of you who like a challenge, please go ahead and implement this feature in your submission but clearly demarcate it in your code and allow this feature to be turned on or off using the command line switch "-x" (where "-x" means to turn on the extended processing of phrasal queries). We will give a small bonus to submissions that achieve this functionality correctly).
  2. Describe how your search engine reacts to long documents and long queries as compared to short documents and queries. Is the normalization you use sufficient to address the problems (see Section 6.4.4 for a hint)? In your judgement, is the lnc.ltc scheme (n.b., not the ranking scheme you were asked to implement) sufficient for retrieving documents from the Reuters-21578 collection?
  3. Do you think zone or field parametric indices would be useful for practical search in the Reuters collection? Note: the Reuters collection does have metadata for each article but the quality of the metadata is not uniform, nor are the metadata classifications uniformly applied (some documents have it, some don't). Hint: for the next Homework #4, we will be using field metadata, so if you want to base Homework #4 on your Homework #3, you're welcomed to start support of this early (although no extra credit will be given if it's right).

Submission Formatting

You are allowed to do this assignment individually or as a team of two. There will be no difference in grading criteria if you do the assignment as a team or individually. For the submission information below, simply replace any mention of a matric number with the two matric numbers concatenated with a separating dash (e.g., U000000X-U000001Y).

For us to grade this assignment in a timely manner, we need you to adhere strictly to the following submission guidelines. They will help me grade the assignment in an appropriate manner. You will be penalized if you do not follow these instructions. Your matric number in all of the following statements should not have any spaces and any letters should be in CAPITALS. You are to turn in the following files:

These files will need to be suitably zipped in a single file called <matric number>.zip. Please use a zip archive and not tar.gz, bzip, rar or cab files. Make sure when the archive unzips that all of the necessary files are found in a directory called <matric number>. Upload the resulting zip file to the IVLE workbin by the due date: 21 March 2014, 11:59:59 pm SGT. There will absolutely be no extensions to the deadline of this assignment. Read the late policy if you're not sure about grade penalties for lateness.

Grading Guidelines

The grading criteria for the assignment is tentatively:

Disclaimer: percentage weights may vary without prior notice.