(for each query)
Information Retrieval and
Extraction
Spring 2018
Homework Webpage
¡@
Homework #1 :Evaluation Metrics for IR
The the query-document relevance information (AssessmentTrainSet.txt) for a set of queries (16 queries) and a collection of 2,265 documents is provided. An IR model is then tested on this query set and save the corresponding ranking results in a file (ResultsTrainSet.txt) . Please evaluate the overall model performance using the following two measures.
1.
Interpolated Recall-Precision Curve:
(for each query)
(overall performance)
2. (Non-interpolated) Mean Average Precision:
, where "non-interpolated average precision" is "average precision at seen relevant documents" introduced in the textbook.
Example 1: Interpolated Recall-Precision Curve
Example 2: (Non-interpolated) Mean Average Precision
mAP=0.63787418
3. Normalized Discounted Cumulated Gain (NDCG) :
Homework #2 : Retrieval Models
A set of text queries (16 queries) and a collection of text documents ( 2,265 documents) is provided, in which each word is represented as a number except that the number "-1" is a delimiter.
Implement an information retrieval system based on the Vector (Space) Model (or Probabilistic Model, Generalized Vector Space Model, Latent Semantic Analysis, Language Model, etc.). The query-document relevance information is in "AssessmentTrainSet.txt". You should evaluated you system with the two measures described in HW#1.
¡@
Homework #3 : Relevance Feedback and Query Expansion
Try to Integrate the function of query expansion and term re-weighting into your retrieval system that has been built in Homework #2. Either (automatic) reference feedback or local analysis can be adopted as the strategy for it, but local analysis is preferred.
Alternatively, you may use the following set of queries (16 short queries) to evaluate your method; it consists the short versions of the original queries employed in Homework #2 and therefore shares the same query-document relevance information that was used in Homework #1 and #2.
¡@
Homework #4: Language Modeling for Information Retrieval
You should try to implement one language model (LM) based retrieval model for information retrieval. The evaluation should also follow the setup of Homework 2. You may refer to the following file (Word_Unigram_Xinhua98Upper.txt) for the background language model.
Alternatively, you may use the following set of queries (16 short queries) to evaluate your method; it consists the short versions of the original queries employed in Homework #2 and therefore shares the same query-document relevance information that was used in Homework #1 and #2.
Furthermore, you can conduct supervised training of the language models (for example, the interpolation weights between the document models and the background model) of the documents using the following resources:
1) listtdt2qry_OutSideforTrain
2) QDRelevanceTDT2_forHMMOutSideTrain
Estimation formula of the interpolation weights (for example, m1):
¡@
¡@