使用Java实施排名融合方法评估,比较和分析基线和排名融合方法

INFS7410 Project – Part 1 – v2
Note: these instructions have been modified on 26/08/2019
Preamble
The due date for this assignment is 29 August 2019 17:00 5 September 2019 17:00, Eastern Australia Standard Time (extended from 29/08).实施排名融合方法
评估，比较和分析基线和排名融合方法
This project is worth 5% of the overall mark for INFS7410. A detailed marking sheet for this assignment is provided at the end of this document.
We recommend that you make an early start on this assignment, and proceed by steps. There are a number of activities you make already tackle, including setting up the pipeline, manipulating the queries, implement some retrieval functions and perform evaluation and analysis. There are some activities you do not know yet how to perform, in particular the implementation of the rank fusion algorithms: this will be the topic of the week 5 lecture and tutorials.
Aim
Project aim: The aim of this project is to implement a number of information retrieval methods, evaluate them and compare them in the context of a real use-case.
Project Part 1 aim
The aim of part 1 is to:
setup the evaluation infrastructure, including collection and index, topics, qrels implement common information retrieval baselines
implement ranking fusion methods
evaluate, compare and analyse baseline and ranking fusion methods
The Information Retrieval Task: Ranking of studies for Systematic Reviews
In this project we will consider the problem of ranking research studies identified as part of a systematic review. Systematic reviews are a widely used method to provide an overview of the current scientific consensus, by bringing together multiple studies in a reliable, transparent way. We will use the CLEF 2017 and 2018 eHealth TAR (task 2) collections. In CLEF TAR 2017, the task we consider is referred to as subtask 1 (and is the only task); in CLEF TAR 2018, the task we consider is referred to as subtask 2. We provide the CLEF 2017 and 2018 TAR task overview papers in the assignment folder in blackboard for your reference. These contain details about the topics, the collection, the task, etc. These details are not necessary to complete the assignment, but nevertheless you may want to know more about this task, its importance, approaches that have been tried, and so on.

The task consists of, given as the starting point the results of the Boolean search created by the researchers undertaking a systematic review, ranking the set of the provided documents (they are PMID – pubmed ID – in the files provided; for each PMID there is an associated title and abstract). The goal is to produce an ordering of the documents such that all the relevant documents are retrieved above the irrelevant ones. This is to be achieved through automatic methods that rank all abstracts, with the goal of retrieving relevant documents as early in the ranking as possible.
There are two datasets to consider in this project. The CLEF 2017 TAR dataset; and the CLEF 2018 TAR dataset. Each dataset consists of material for training, and. material for testing the developed information retrieval methods.
What we provide you with
We provide:
for each dataset, a list of topics to be used for training. Each topic is organised into a file. Each topic contains a title and a Boolean query.
for each dataset, a list of topics to be used for testing. Each topic is organised into a file. Each topic contains a title and a Boolean query.
each topic file (both those for training and those for testing), includes a list of retrieved documents in the form of their PMIDs: these are the documents that you have to rank. Take note: you do not need to perform the retrieval from scratch (i.e. execute the query against the whole index); instead you need to rank (order) the provided documents.
for each dataset, and for each train and test partition, a qrels file, containing relevance assessments for the documents to be ranked. This is to be used for evaluation.
for each dataset, and for test partitions, a set of runs from retrieval systems that participated to CLEF 2017/2018 to be considered for fusion.
a Terrier index of the entire Pubmed collection. This index has been produced using the Terrier stopword list and Porter stemmer.
a Java Maven project that contains the Terrier dependencies and a skeleton code to give you a start. NOTE: Tip #1 provides you with a restructured skeleton code to make the processing of queries more efficient.
a template for your project report.
What you need to produce
You need to produce:
correct implementations of the methods required by this project specifications
correct evaluation, analysis and comparison of the evaluated methods, written up into a report following the provided template
a project report that, following the provided template, details: an explanation of the retrieval methods used, an explanation of the evaluation settings followed, the evaluation of results (as described above), inclusive of analysis, a discussion of the findings.
Required methods to implement
In part 1 of the project you are required to implement the following retrieval methods:
1. TF-IDF: you can create your own implementation using the Terrier API to extract index

statistics, or use the implementation available through the Terrier API
2. BM25: you can create your own implementation using the Terrier API to extract index
statistics, or use the implementation available through the Terrier API
3. The ranking fusion method Borda; you need to create your own implementation of this
4. The ranking fusion method CombSUM; you need to create your own implementation of this
5. The ranking fusion method CombMNZ; you need to create your own implementation of this
We strongly reccommend you use the provided Maven project to implement these methods. You should have already attempted many of the implementations above as part of the tutorial exercises.
In the report, detail how the methods were implemented, i.e. (i) which formula you implemented, (ii) if you did your own implementation or levereged Terrier’s ones (for TF-IDF and BM25).
For ranking fusion methods, consider to fuse the runs from previous participants from CLEF 2017/2018 we provide, and the TF-IDF and the BM25 runs you will produce.
What queries to use
We ask you to consider two types of queries for each topic (the second type is optional and attracts bonus points):
1. for each topic, a query created from the topic title. For example, consider the example (partial) topic listed below: the query will be Rapid diagnostic tests for diagnosing uncomplicated P. falciparum malaria in endemic countries (you may consider performing text processing).
2. (OPTIONAL: 2% bonus if done) for each topic, a query created from the Boolean query associated with the topic. This Boolean query will be made up of the terms that appear in the query, but will ignore any operator (e.g., will ignore and , or, Exp , / , etc.) and field restrictions (e.g., .ti , .ab , .ti,ab , etc.). Note that some keywords in the Boolean query have been manually stemmed, e.g. diagnos* in the example topic below. As part of the query creation process, we ask you to use the entrez API. For documentation on the entrez esearch API, please refer to the Entrez Programming Utilities Help reference available at: https://www.ncbi.nlm.nih.gov/books/NBK25499/#chapter4.ESearch. Example usage can be found at the following URL: https://eutils.ncbi.nlm.nih.gov/entrez/eutils/esearch.fcgi? db=pubmed&term=diagnos*. Note the terms in the TranslationStack field. These are the terms you would use to replace diagnosis* and therefore concatenate to form the query (along with the other terms).

Title: Rapid diagnostic tests for diagnosing uncomplicated P. falciparum
malaria in endemic countries
Query:
1. Exp Malaria/
2. Exp Plasmodium/
3. Malaria.ti,ab
4. 1or2or3
5. Exp Reagent kits, diagnostic/ 6. rapid diagnos* test*.ti,ab
7. RDT.ti,ab
8. Dipstick*.ti,ab
Above: example topic file
More on the Entrez API
The Entrez API provides access to the Pubmed search functionalities. In this part of the project we will not use this API for retrieval. However, it also provide some additional method. One in particular is useful for expanding terms in the Boolean query that have been “wildcarded” (manually stemmed): the TranslationStack . We have shown you above an example of how to obtain the output of the TranslationStack for a stem term. You will have to use this method for all terms in the Boolean query that contain the wildcard operator * . Practically, you will need to make a call to this API by constructing an appropriate URL, then request that URL, and finally parsing the response to obtain the list of index terms to use to substitute the wildcarded term from the boolean query for inclusion in your text query. Note that it is likely that one wildcarded term will give rise to many terms you will add to your query.
Tips on making query processing efficient
A number of tips have been provided in Blackboard to make the execution of queries more efficient. Please consider these tips to reduce the execution time of the experiments.
Required evaluation to perform
In part 1 of the project you are required to perform the following evaluation:
1. For all methods, train on the training set for the 2017 topics (train here means you use this data to tune any parameter of a retrieval model, e.g. and for BM25, runs to be considered for the rank fusion methods, etc.) and test on the testing set for the 2017 topics (using the parameter values you selected from the training set). Report the results of every method on the training and on the testing set, separately, into one table. Perform statistical significance analysis across the results of the methods.
2. Comment on the results reported in the previous table by comparing the methods on the 2017 dataset.
3. For all methods, train on the training set for the 2018 topics (train here means you use this data to tune any parameter of a retrieval model, e.g. and for BM25, runs to be

considered for the rank fusion methods, etc.) and test on the testing set for the 2018 topics (using the parameter values you selected from the training set). Report the results of every method on the training and on the testing set, separately, into one table. Perform statistical significance analysis across the results of the methods.
4. Comment on the results reported in the previous table by comparing the methods on the 2018 dataset.
5. Perform a topic-by-topic gains/losses analysis for both 2017 and 2018 results on the testing datasets, by considering as baseline BM25, and as comparison each of TF-IDF, Borda, CombSUM and CombMNZ.
6. Comment on trends and differences observed when comparing the findings from 2017 and 2018 results. Is there a method that consistently outperform the others?
7. Provide insights of when ranking fusion works, and when it does not, e.g. with respect to runs to be considered in the fusion process, queries, etc.
In terms of evaluation measures, evaluate the retrieval methods with respect to mean average precision (MAP) using trec_eval . Remember to set the cut-off value ( -M , i.e. the maximum number of documents per topic to use in evaluation) to the number of documents to be re- ranked for each of the queries. Using trec_eval , also compute Rprecision (Rprec), which is the precision after R documents have been retrieved (by default, R is the total number of relevant docs for the topic).
For all statistical significance analysis, use paired t-test; distinguish between p<0.05 and p<0.01.
Perform the above analysis for: 1. queries created from topic files using the topic title; 2. (OPTIONAL) queries created from the topic files using the Boolean queries. Finish your analysis by comparing the effectiveness difference between the methods using topic titles and those using queries extracted from the Boolean queries (OPTIONAL: to do only if you do consider Boolean queries and want to obtain the bonus points).
How to submit
You will have to submit 3 files:
1. the report, formatted according to the provided template, saved as PDF or MS Word document
2. a zip file containing all the runs (result files) you have created for the implemented methods
3. a zip file containing all the code to re-run your experiments. You do not need to include in
this zip file the runs we have given to you. You may need to include additional files e.g. if you manually process the topic files into an intermediate format (rather than automatically process them from the files we provide you), so that we can re-run your experiments to confirm your results and implementation.
All items need to be submitted via the relevant Turnitin link in the INFS7410 Blackboard site, by 29 August 2019 17:00, Eastern Australia Standard Time, unless you have been given an extension (according to UQ policy), before the due date of the assignment.

INFS 7410 Project Part 1 – Marking Sheet – v2
Criterion
%
7 100%
4 50%
FAIL 1 0%
IMPLEMENTATION The ability to:
• Understand
implement and execute common IR baseline
• Understand implement and
execute rank
fusion methods
• Perform text
processing
2
• Correctly implements the specified baselines and the rank fusion methods
• Implemented methods to deal with title queries
• (OPTIONAL:) Implemented methods deal with Boolean queries, and wildcards are appropriately handled via expansion to possible forms using provided API (2% bonus)
• Correctly implements the specified baselines and the rank fusion methods
• No implementation
• Implements only baselines, but not
the rank fusion methods
EVALUATION The ability to:
• Empirically evaluate
and compare IR
methods
• Analyse the results of
empirical IR
evaluation
• Analyse the statistical
significance difference between IR methods’ effectiveness
2
• Correct empirical evaluation has been performed
• Uses all required evaluation measures
• Correct handling of the tuning regime (train/test)
• Reports all results for the
provided query sets into
appropriate tables
• Provides graphical analysis
of results on a query-by- query basis using appropriate gain-loss plots
• Provides correct statistical significance analysis within the result table; and correctly describes the statistical analysis performed
• Provides a written understanding and discussion of the results with respect to the methods
• Provides examples of where fusion works, and were it does not, and why, e.g., discussion with respect to queries, runs.
• Correct empirical evaluation has been performed
• Uses all required evaluation measures
• Correct handling of the tuning regime (train/test)
• Reports all results for the provided query sets into appropriate tables
• Provides graphical analysis of results on a query-by-query basis using appropriate gain-loss plots
• Does not perform statistical significance analysis, or errors are present in the analysis
• No or only partial empirical evaluation has been conducted, e.g. only on a topic set, or a subset of topics
• Only report a partial set of evaluation measures
• Fails to correctly handle training and testing partitions, e.g. train on test, reports only overall results
WRITE UP
Binary score: 0/1
The ability to:
• use fluent
language with correct grammar, spelling and punctuation
• use appropriate paragraph,
sentence
structure
• use appropriate
style and tone of
writing
• produce a
professionally presented document, according to the provided template
1
• Structure of the document is appropriate and meets expectations
• Clarity promoted by consistent use of standard grammar, spelling and punctuation
• Sentences are coherent
• Paragraph structure
effectively developed
• Fluent, professional style
and tone of writing.
• No proof reading errors
• Polished professional
appearance
• Written expression and
presentation are incoherent, with little or no structure, well below
required standard
• Structure of the document is not appropriate and does not meet expectations
• Meaning unclear as grammar and/or spelling contain frequent errors.
• Disorganised or incoherent writing.