#informationretrieval — Public Fediverse posts

https://www.europesays.com/people/62875/ Meta Is Building an AI Version of Mark Zuckerberg for Employees #DigitalClone #InformationRetrieval #MarkZuckerberg

RE: https://researchbuzz.masto.host/@mottg/116474491390856287

Nice, the state of Information Retrieval in 2026 by Mohan Krishna. Lots of interesting references and thoughts, if you're into leaderboards and state-of-the-art performance on benchmark test collections. #InformationRetrieval

RE: https://researchbuzz.masto.host/@mottg/116474491390856287

Nice, the state of Information Retrieval in 2026 by Mohan Krishna. Lots of interesting references and thoughts, if you're into leaderboards and state-of-the-art performance on benchmark test collections. #InformationRetrieval

RE: https://researchbuzz.masto.host/@mottg/116474491390856287

Nice, the state of Information Retrieval in 2026 by Mohan Krishna. Lots of interesting references and thoughts, if you're into leaderboards and state-of-the-art performance on benchmark test collections. #InformationRetrieval

RE: https://researchbuzz.masto.host/@mottg/116474491390856287

Nice, the state of Information Retrieval in 2026 by Mohan Krishna. Lots of interesting references and thoughts, if you're into leaderboards and state-of-the-art performance on benchmark test collections. #InformationRetrieval

RE: https://researchbuzz.masto.host/@mottg/116474491390856287

Nice, the state of Information Retrieval in 2026 by Mohan Krishna. Lots of interesting references and thoughts, if you're into leaderboards and state-of-the-art performance on benchmark test collections. #InformationRetrieval

Published at #IRRJ: "Simple Techniques for Efficient Top-k Batch Query Processing" by Zhixuan Li and Joel Mackenzie. #BatchProcessing, #Caching, #DynamicPruning, #EfficientRetrieval, #InformationRetrieval

https://doi.org/10.54195/irrj.23893

Published at #IRRJ: "Simple Techniques for Efficient Top-k Batch Query Processing" by Zhixuan Li and Joel Mackenzie. #BatchProcessing, #Caching, #DynamicPruning, #EfficientRetrieval, #InformationRetrieval

https://doi.org/10.54195/irrj.23893

Published at #IRRJ: "Simple Techniques for Efficient Top-k Batch Query Processing" by Zhixuan Li and Joel Mackenzie. #BatchProcessing, #Caching, #DynamicPruning, #EfficientRetrieval, #InformationRetrieval

https://doi.org/10.54195/irrj.23893

Published at #IRRJ: "Simple Techniques for Efficient Top-k Batch Query Processing" by Zhixuan Li and Joel Mackenzie. #BatchProcessing, #Caching, #DynamicPruning, #EfficientRetrieval, #InformationRetrieval

https://doi.org/10.54195/irrj.23893

Published at #IRRJ: "Simple Techniques for Efficient Top-k Batch Query Processing" by Zhixuan Li and Joel Mackenzie. #BatchProcessing, #Caching, #DynamicPruning, #EfficientRetrieval, #InformationRetrieval

https://doi.org/10.54195/irrj.23893

Looking back at #ECIR2026: Jaap Kamps presented the #IRRJ paper "Effectiveness of In-Context Learning for Due Diligence": https://doi.org/10.54195/irrj.22626 #InformationRetrieval

Looking back at #ECIR2026: Jaap Kamps presented the #IRRJ paper "Effectiveness of In-Context Learning for Due Diligence": https://doi.org/10.54195/irrj.22626 #InformationRetrieval

Looking back at #ECIR2026: Jaap Kamps presented the #IRRJ paper "Effectiveness of In-Context Learning for Due Diligence": https://doi.org/10.54195/irrj.22626 #InformationRetrieval

Looking back at #ECIR2026: Jaap Kamps presented the #IRRJ paper "Effectiveness of In-Context Learning for Due Diligence": https://doi.org/10.54195/irrj.22626 #InformationRetrieval

Looking back at #ECIR2026: Jaap Kamps presented the #IRRJ paper "Effectiveness of In-Context Learning for Due Diligence": https://doi.org/10.54195/irrj.22626 #InformationRetrieval

A practical look at how search indexing is evolving to hybrid retrieval systems that support semantic search, vector search, and AI-driven query understanding. https://hackernoon.com/from-inverted-indexes-to-hybrid-retrieval-rethinking-search-architecture #informationretrieval

A practical look at how search indexing is evolving to hybrid retrieval systems that support semantic search, vector search, and AI-driven query understanding. https://hackernoon.com/from-inverted-indexes-to-hybrid-retrieval-rethinking-search-architecture #informationretrieval

A practical look at how search indexing is evolving to hybrid retrieval systems that support semantic search, vector search, and AI-driven query understanding. https://hackernoon.com/from-inverted-indexes-to-hybrid-retrieval-rethinking-search-architecture #informationretrieval

A practical look at how search indexing is evolving to hybrid retrieval systems that support semantic search, vector search, and AI-driven query understanding. https://hackernoon.com/from-inverted-indexes-to-hybrid-retrieval-rethinking-search-architecture #informationretrieval

A practical look at how search indexing is evolving to hybrid retrieval systems that support semantic search, vector search, and AI-driven query understanding. https://hackernoon.com/from-inverted-indexes-to-hybrid-retrieval-rethinking-search-architecture #informationretrieval

Well i finally did it. I just released my test dataset for AI Evaluation. Its a simulated company, represented by 60,000 documents, the readme in the image explains it all ... If you are interested, its at https://codeberg.org/Lorenz_Systems/Company_Sim.git

#EUAIAct #DigitalSovereignty #SovereignCloud #FOSS #FLOSS #Codeberg #Forgejo #OpenSource #DataGovernance #Auditability #ForensicAI #EUTech #PrivacyByDesign #InformationRetrieval #KnowledgeManagement #DeterministicAI #EUPL

Well i finally did it. I just released my test dataset for AI Evaluation. Its a simulated company, represented by 60,000 documents, the readme in the image explains it all ... If you are interested, its at https://codeberg.org/Lorenz_Systems/Company_Sim.git

#EUAIAct #DigitalSovereignty #SovereignCloud #FOSS #FLOSS #Codeberg #Forgejo #OpenSource #DataGovernance #Auditability #ForensicAI #EUTech #PrivacyByDesign #InformationRetrieval #KnowledgeManagement #DeterministicAI #EUPL

Well i finally did it. I just released my test dataset for AI Evaluation. Its a simulated company, represented by 60,000 documents, the readme in the image explains it all ... If you are interested, its at https://codeberg.org/Lorenz_Systems/Company_Sim.git

#EUAIAct #DigitalSovereignty #SovereignCloud #FOSS #FLOSS #Codeberg #Forgejo #OpenSource #DataGovernance #Auditability #ForensicAI #EUTech #PrivacyByDesign #InformationRetrieval #KnowledgeManagement #DeterministicAI #EUPL

@marlinz #Informationretrieval is depending on the current task and situation.

Sometimes #navigation is the most efficient one, sometimes it's #search. (my #TagTrees are a combination of both) And then there is teleporting, bookmarks/favorites, ...

Unfortunately, most people don't have the knowledge and experience to use multiple methods and decide which method to choose in a specific situation. 😞

#PIM

@marlinz #Informationretrieval is depending on the current task and situation.

Sometimes #navigation is the most efficient one, sometimes it's #search. (my #TagTrees are a combination of both) And then there is teleporting, bookmarks/favorites, ...

Unfortunately, most people don't have the knowledge and experience to use multiple methods and decide which method to choose in a specific situation. 😞

#PIM

@marlinz #Informationretrieval is depending on the current task and situation.

Sometimes #navigation is the most efficient one, sometimes it's #search. (my #TagTrees are a combination of both) And then there is teleporting, bookmarks/favorites, ...

Unfortunately, most people don't have the knowledge and experience to use multiple methods and decide which method to choose in a specific situation. 😞

#PIM

@marlinz #Informationretrieval is depending on the current task and situation.

Sometimes #navigation is the most efficient one, sometimes it's #search. (my #TagTrees are a combination of both) And then there is teleporting, bookmarks/favorites, ...

Unfortunately, most people don't have the knowledge and experience to use multiple methods and decide which method to choose in a specific situation. 😞

#PIM

@marlinz #Informationretrieval is depending on the current task and situation.

Sometimes #navigation is the most efficient one, sometimes it's #search. (my #TagTrees are a combination of both) And then there is teleporting, bookmarks/favorites, ...

Unfortunately, most people don't have the knowledge and experience to use multiple methods and decide which method to choose in a specific situation. 😞

#PIM

Retrieval-Augmented Generation: When Retrieval Stops Short of an Answer

Part 1: Understanding the Problem RAG Solves

RAG is commonly referred to as a means to enhance language models. In effect, it performs the reverse. RAG controls what the model can see. It can only work with what retrieval gives it. The limitation is what makes RAG valuable.

This article explains what that means, and why it matters.

A Simple, Familiar Dataset

To keep the discussion simple, consider a dataset of movie reviews.

Each record contains:

• A short review
• A rating
• A genre
• The movie’s release year

This is kind of data is public, familiar, and easy to reason about. More importantly, it contains repeated opinions expressed in different ways, which makes it ideal for understanding where retrieval alone falls short.

Where Retrieval Naturally Stops

We already know retrieval. Search engines, databases, semantic pipelines, we have been doing this for years. RAG doesn’t reinvent that, it uses the same.

What RAG adds is two things.

First, the query itself can now be natural language, or we can additionally use intermediate representation of query in a structured json. This feeds into hybrid search that combines semantic and keyword matching. Second, and more importantly, we don’t stop at retrieval.

Once this information is retrieved, we interpret it. We augment it. Instead of throwing a list of raw results at the user and saying “here, figure it out”, the system takes responsibility for making that information meaningful and easy to consume. That’s the shift RAG introduces, and it’s a powerful one.

Example 1: Aggregation Within a Single Movie

A user may ask a question: “What do people generally like and dislike about this movie?”

A semantic search engine pulls up multiple relevant reviews:

• “The performances were strong, but the pacing felt slow.”
• “Great acting and visuals, though the story dragged.”
• “Loved the cinematography, but the movie could have been shorter.”

From a retrieval perspective, this is a success. The reviews are clearly relevant. But the user’s question is still unanswered.

The system returns examples, not an answer. The user must still read the reviews, identify recurring ideas, and summarize the result mentally. Retrieval has surfaced the data, but interpretation is still manual.

The Shift Introduced by RAG

RAG does not change how information is retrieved, but it doesnt throw out the raw information either. It changes what the system does with retrieved information. Instead of returning raw results and leaving the interpretation to the user, the system takes responsibility for aggregating and structuring what has been retrieved.

A Minimal Mental Model

Retrieval only:

Query → retrieve relevant reviews → return a list → human interprets

Retrieval with RAG:

Query → retrieve relevant reviews → language model aggregates patterns → system answers in plain English

The retrieval step is unchanged. The only addition is a generation step that works over the retrieved context.

Let’s look at Example 1 again.

With the same retrieved reviews, the system gives the language model a distinct instruction: respond to the question solely using the provided text.

A possible answer:

“Generally, viewers praised the acting and the cinematography. But they often mention slow pacing and dragged storyline.”

This response is not retrieved directly. It is built by pooling repeated signals found across reviews. RAG’s first contribution is the step from examples to a structured answer, which is the step that is most visible..

Beyond Summarization: Aggregation Across Many Movies

RAG becomes more powerful when the question requires patterns across many records, not just interpretation of one.

Example 2: Aggregation with Counts

Now consider this question: “Across movies released in 2025, what themes are most frequently praised and criticized?”

The retrieval step acquires reviews for movies released in the year 2025. The generation step identifies recurring themes, examines the frequency of their occurrence and presents the output in structured format.

Example Output:

Frequently praised:

Strong performances (18/25 reviews)
Visual effects and cinematography (15/25)
Original premise (9/25)

Frequently criticized:

Slow or uneven pacing (14/25)
Predictable storyline (11/25)
Overlong runtime (7/25)

No fresh data will be fetched. No new information is added. This is genuine aggregation that is based on retrieved evidence.

Generation Is Limited by Retrieved Context

The RAG system works based on available information. The answer of the model must be generated using only the bounded context at retrieval step.

If the retrieval is not complete or biased, then the aggregation will also be incomplete. This limitation simplifies reasoning about the system and debugging it because outputs can always be traced back to retrieved evidence.

What This Minimal RAG System Can and Cannot Do

A minimum RAG setup is capable of achieving opinion summarization on a single entity, pattern aggregation on many entities, frequency signal surfacing by counting, and cognitive effort reduction on the user. None of these processes requires models to be retrained or data formats to be altered.

It cannot confirm factual accuracy, deduce intent beyond retrieved evidence, or compensate for absent or biased data. These aren’t defects. They outline the responsibilities within the framework of the system.

Importance Of Starting Minimal

RAG architectures can rapidly evolve into complicated designs comprising filters, rankers and agent based workflows. An initial ‘minimal’ design makes our core idea clear: retrieval decides what information is available; generation determines how that information is aggregated and expressed.

Once this border is clear, more sophisticated techniques can be introduced intentionally rather than randomly.

Final Thought

Search systems help find information. Retrieval Augmented Generation(RAG) assists systems in interpreting and aggregating data.

In the next section, we’ll explore how retrieval works, and where it silently fails.

👥 Justus-Jonas Erker (UKP Lab/Technische Universität Darmstadt), Nils Reimers (Cohere), Iryna Gurevych (UKP Lab/Technische Universität Darmstadt)

See you at Hashtag#EACL2026 in Rabat 🕌!

#UKPLab #NLP #NLProc #InformationRetrieval #DenseRetrieval #MultiHop #FactChecking #QuestionAnswering #RAG

👥 Justus-Jonas Erker (UKP Lab/Technische Universität Darmstadt), Nils Reimers (Cohere), Iryna Gurevych (UKP Lab/Technische Universität Darmstadt)

See you at Hashtag#EACL2026 in Rabat 🕌!

#UKPLab #NLP #NLProc #InformationRetrieval #DenseRetrieval #MultiHop #FactChecking #QuestionAnswering #RAG

👥 Justus-Jonas Erker (UKP Lab/Technische Universität Darmstadt), Nils Reimers (Cohere), Iryna Gurevych (UKP Lab/Technische Universität Darmstadt)

See you at Hashtag#EACL2026 in Rabat 🕌!

#UKPLab #NLP #NLProc #InformationRetrieval #DenseRetrieval #MultiHop #FactChecking #QuestionAnswering #RAG

👥 Justus-Jonas Erker (UKP Lab/Technische Universität Darmstadt), Nils Reimers (Cohere), Iryna Gurevych (UKP Lab/Technische Universität Darmstadt)

See you at Hashtag#EACL2026 in Rabat 🕌!

#UKPLab #NLP #NLProc #InformationRetrieval #DenseRetrieval #MultiHop #FactChecking #QuestionAnswering #RAG

FYI: AI-Powered Search: Explained in Simple Terms #shorts: What is AI-powered search? It's where artificial intelligence meets information retrieval. AI allows computers to behave like people, using machine learning and data science to improve search results. It's not just about chatbots; it's about smarter search. #AIsearch #machinelearning #datavisualization #informationretrieval https://www.youtube.com/shorts/tflRHOGMj6Q

FYI: AI-Powered Search: Explained in Simple Terms #shorts: What is AI-powered search? It's where artificial intelligence meets information retrieval. AI allows computers to behave like people, using machine learning and data science to improve search results. It's not just about chatbots; it's about smarter search. #AIsearch #machinelearning #datavisualization #informationretrieval https://www.youtube.com/shorts/tflRHOGMj6Q

FYI: AI-Powered Search: Explained in Simple Terms #shorts: What is AI-powered search? It's where artificial intelligence meets information retrieval. AI allows computers to behave like people, using machine learning and data science to improve search results. It's not just about chatbots; it's about smarter search. #AIsearch #machinelearning #datavisualization #informationretrieval https://www.youtube.com/shorts/tflRHOGMj6Q

FYI: AI-Powered Search: Explained in Simple Terms #shorts: What is AI-powered search? It's where artificial intelligence meets information retrieval. AI allows computers to behave like people, using machine learning and data science to improve search results. It's not just about chatbots; it's about smarter search. #AIsearch #machinelearning #datavisualization #informationretrieval https://www.youtube.com/shorts/tflRHOGMj6Q

Big congratulations to all authors! 🚀

#ICLR2026 #MachineLearning #AIResearch #RepresentationLearning #InformationRetrieval #DenseRetrieval #SelfSupervisedLearning #LanguageModels #NLP #UKPLab #ICLR

Big congratulations to all authors! 🚀

#ICLR2026 #MachineLearning #AIResearch #RepresentationLearning #InformationRetrieval #DenseRetrieval #SelfSupervisedLearning #LanguageModels #NLP #UKPLab #ICLR

Big congratulations to all authors! 🚀

#ICLR2026 #MachineLearning #AIResearch #RepresentationLearning #InformationRetrieval #DenseRetrieval #SelfSupervisedLearning #LanguageModels #NLP #UKPLab #ICLR

Big congratulations to all authors! 🚀

#ICLR2026 #MachineLearning #AIResearch #RepresentationLearning #InformationRetrieval #DenseRetrieval #SelfSupervisedLearning #LanguageModels #NLP #UKPLab #ICLR

ICYMI: AI-Powered Search: Explained in Simple Terms #shorts: What is AI-powered search? It's where artificial intelligence meets information retrieval. AI allows computers to behave like people, using machine learning and data science to improve search results. It's not just about chatbots; it's about smarter search. #AIsearch #machinelearning #datavisualization #informationretrieval https://www.youtube.com/shorts/tflRHOGMj6Q

ICYMI: AI-Powered Search: Explained in Simple Terms #shorts: What is AI-powered search? It's where artificial intelligence meets information retrieval. AI allows computers to behave like people, using machine learning and data science to improve search results. It's not just about chatbots; it's about smarter search. #AIsearch #machinelearning #datavisualization #informationretrieval https://www.youtube.com/shorts/tflRHOGMj6Q

ICYMI: AI-Powered Search: Explained in Simple Terms #shorts: What is AI-powered search? It's where artificial intelligence meets information retrieval. AI allows computers to behave like people, using machine learning and data science to improve search results. It's not just about chatbots; it's about smarter search. #AIsearch #machinelearning #datavisualization #informationretrieval https://www.youtube.com/shorts/tflRHOGMj6Q

ICYMI: AI-Powered Search: Explained in Simple Terms #shorts: What is AI-powered search? It's where artificial intelligence meets information retrieval. AI allows computers to behave like people, using machine learning and data science to improve search results. It's not just about chatbots; it's about smarter search. #AIsearch #machinelearning #datavisualization #informationretrieval https://www.youtube.com/shorts/tflRHOGMj6Q

RAG-системы: что это такое, принципы работы, архитектура и ограничения

Retrieval-Augmented Generation (RAG) всё чаще упоминается в контексте LLM и всё чаще фигурирует в требованиях к разработчикам, но за этим термином обычно скрывается довольно размытое представление о том, как такие системы реально устроены. В этой статье я разбираю RAG как архитектурный подход: зачем он вообще появился, какие задачи решает, как выглядит базовый пайплайн от данных до ответа модели и где на практике чаще всего возникают проблемы.

https://habr.com/ru/articles/989000/

#rag #llm #retrieval #nlp #embeddings #semanticsearch #informationretrieval

RAG-системы: что это такое, принципы работы, архитектура и ограничения

Retrieval-Augmented Generation (RAG) всё чаще упоминается в контексте LLM и всё чаще фигурирует в требованиях к разработчикам, но за этим термином обычно скрывается довольно размытое представление о том, как такие системы реально устроены. В этой статье я разбираю RAG как архитектурный подход: зачем он вообще появился, какие задачи решает, как выглядит базовый пайплайн от данных до ответа модели и где на практике чаще всего возникают проблемы.

https://habr.com/ru/articles/989000/

#rag #llm #retrieval #nlp #embeddings #semanticsearch #informationretrieval

RAG-системы: что это такое, принципы работы, архитектура и ограничения

Retrieval-Augmented Generation (RAG) всё чаще упоминается в контексте LLM и всё чаще фигурирует в требованиях к разработчикам, но за этим термином обычно скрывается довольно размытое представление о том, как такие системы реально устроены. В этой статье я разбираю RAG как архитектурный подход: зачем он вообще появился, какие задачи решает, как выглядит базовый пайплайн от данных до ответа модели и где на практике чаще всего возникают проблемы.

https://habr.com/ru/articles/989000/

#rag #llm #retrieval #nlp #embeddings #semanticsearch #informationretrieval

RAG-системы: что это такое, принципы работы, архитектура и ограничения

Retrieval-Augmented Generation (RAG) всё чаще упоминается в контексте LLM и всё чаще фигурирует в требованиях к разработчикам, но за этим термином обычно скрывается довольно размытое представление о том, как такие системы реально устроены. В этой статье я разбираю RAG как архитектурный подход: зачем он вообще появился, какие задачи решает, как выглядит базовый пайплайн от данных до ответа модели и где на практике чаще всего возникают проблемы.

https://habr.com/ru/articles/989000/

#rag #llm #retrieval #nlp #embeddings #semanticsearch #informationretrieval