Discover research smarter and faster
Type your thesis title or topic. LIRAS reads it the way a researcher would and points you to the most relevant studies.
Searches 942 ISU-Cauayan theses and 28,364 arXiv papers
Isabela State University
College of Computing Studies, Information and Communication Technology
ISU-CYN Library
Search once, read what matters
-
Hybrid recommendation engine
Keyword matching and meaning-based matching, blended into one ranked score.
-
ML-powered recommendation
Word embeddings read meaning, not just spelling.
-
Keyword search analysis
Titles that share your exact terms always show up first.
-
Synonym and related-term matching
Finds papers that mean the same thing, even when phrased differently.
-
Title similarity matching
Compares your topic to thousands of thesis titles and shows how closely each one relates.
-
Fast paper discovery
Results in seconds, so your hours go to writing.
From one title to your reading list
Four stages, in the order the engine runs them.
Enter or search
Type a working thesis title or research topic. One line is enough.
NLP preprocessing
Your title is lowercased, cleaned and stemmed so it can be compared fairly.
Smart matching
Exact words and meaning are checked at once, then blended into one score.
Ranked papers
The most relevant papers come first, each with its match score.
29,306
research titles, searchable from one box
Mean Absolute Error
12%
On average, a LIRAS similarity score is 12% away from the experts' rating.
Root Mean Square Error
15%
Like MAE, but large misses count more. Staying this low means big misses are rare.
Pearson correlation
89%
LIRAS scores rise and fall with the experts' scores. 100% would be a perfect match.
Spearman correlation
91%
LIRAS ranks titles in nearly the same order the experts do. 100% would be identical.
Frequently asked questions
You give it a thesis title or research topic, and it returns the most relevant research papers from a combined library of 29,306 titles, ranked by a hybrid score that blends word-level and meaning-level similarity.
TF-IDF is precise about exact terminology but blind to synonyms; Word2Vec understands meaning but can drift from specific keywords. Fused at α = 0.78, the hybrid outperformed either model alone in the experiments, tracking expert judgment most closely.
The optimal model achieved a mean absolute error of 12% and a Pearson correlation of 89%.
Students starting a literature review, faculty checking whether a proposed title overlaps with past work, and librarians who want a smarter front door to the university archive. No technical background needed.