Tiklso link

Need java expert for information retrieval project - Upwork

Biudžetas 501-1000 Eur
Sukurta: 2019-10-14
Baigiasi: 2019-10-21
Siūlo: Nėra
Apibūdinimas: this is the requirement

Required methods to implement

In part 3 of the project you are required to implement a method that you believe will provide results better than the baselines(BM25,TF-IDF) considered in part 1. For the method you implement, consider the re-ranking task, that is, use your method only to re-rank the documents that are provided in the topic file, and do not insert into the ranking any document that was not initially retrieved and provided in the topic file. We provide two options to tackle this part of the project. You can choose to follow either of the two options.

Option A: your own choice Implement a method of your choice, including proposing a radically new method. For example you may consider a query expansion technique (be considerate about the amount of expansion though, and the time required to run large queries), a learning to rank approach, or a different retrieval model, among other possible choices. Your method can consider using relevance information in an iterative fashion (e.g., see below for choice B), but only for a document that has been already placed into an iterative ranking mechanism.

Option B: relevance feedback Implement the following method, making decisions on your own as you see best fit given what you have learned in the course, with respect to the details of the implementation and of the settings. The method is Relevance Feedback, using either of (1) BM25 (the full formula with RSJ weight), (2) Binary Independence Model (where you would use relevance feedback information to set the relevant statistics), (3) The Rocchio Algorithm. To implement the relevance feedback method, do the following: (1) start from the BM25 baseline you consider in your experiments; (2) the first document BM25 ranks, is the first document your system ranks; (3) for every subsequent rank i+1, to decide which document should be placed, consider the documents at rank 1 to i and acquire their true relevance labels from the qrels; then compute the scores of the documents you have not ranked yet according to the Relevance Feedback method you have chosen; (4) place at rank i the document with the highest score identified at point 3; (5) continue with this process until all documents have been re-ranked. Note that once a document has been placed at rank i, you should not change its position depending on its relevance label. In other words, the process describe above consists of presenting the first document from your baseline to an imaginary user, then gather its relevance assessment for that document. Then, identify the next document to present to the user, by considering the feedback given so far. Once the next document is displayed, feedback is again gathered and used, along with all feedback provided up until now, to identify the next document to show to the user. This proceeds in an iterative way.

This job was posted from a mobile device, so please pardon any typos or any missing details.

Budget: $70

Posted On: October 14, 2019 03:00 UTC
Category: Engineering & Architecture > Other - Engineering & Architecture

Skills: Java
Country: Pakistan

click to apply

Darbo Tipas(ai):
  • PHP
  • CSS
Duomenų Bazė:
Operacinė Sistema: Linux
Siūlymų Skaičius: 0
Siūlosi Žinutės Kaina Trukmė Įvertinimas Informacija