Semantic Search and LLM: What Can They Do?

2023-08-03

Stack Overflow states, "Semantic search and LLM are inseparable, like cookies and milk." Large language models like ChatGPT cannot understand what you say like your friends do. This is because these models have contextual issues. Since 2021, the focus has shifted from backlinks and keywords to understanding intent, behavior, and the semantics behind them. Generative search has opened up a new field of semantic search, and LLM with sequence generation capabilities is the ideal choice for this task. A few days ago, due to increasing page traffic, the question and answer platform Stack Overflow decided to switch to semantic search. In a blog announcement, the company states, "Semantic search and LLM are inseparable, like cookies and milk." This company, founded in 2008, has been using Microsoft SQL's full-text search and Elasticsearch for the past few years. "But even with the use of state-of-the-art algorithms, search encounters some serious problems," the blog writes. The blog lists the reasons for transitioning to semantic search. Firstly, keyword search is very rigid. If there are spelling errors or the use of synonyms, the desired content cannot be found unless the index is processed. If you enter a question as if seeking help, the likelihood of not finding any matching documents is very high. Secondly, keyword search is not intuitive for those who use specific punctuation and Boolean operators for searching. Language models have extraordinary capabilities: they can not only discover relevant information but also respond in natural language, providing a human-like conversational experience during the search process. This LLM feature has been proven beneficial for chatbots and question-answering systems. In layman's terms, semantic search is about understanding the meaning and intent behind queries in a human-like way. Therefore, it can provide accurate search results relevant to the context. Additionally, the integration of LLM and text embeddings allows for faster document retrieval, significantly reducing user search time. Stack Overflow states that its "mission is simple: accuracy and attribution." Although the results generated by large language models (LLMs) have unknown sources, the company has taken steps to explicitly attribute the questions and answers used in its retrieval augmented generation (RAG) LLM replies. Last year, Spotify, the undisputed leader in music services and podcasts, used semantic search to improve the platform's user experience. This included leveraging their integrated podcast creation application, Anchor, for semantic search to enhance the podcast API and use natural language for podcast search functionality. Previously, users had to rely on keyword matching to discover podcasts of interest. However, with the introduction of semantic search, the search experience became more like a conversation with a friend, resulting in significant improvements in search results. This new approach allows for searching based on the meaning of words and sentences, rather than just specific terms, resulting in a more accurate podcast search experience. Google's goal is to become a fully semantic search engine. It is worth noting that all of its major innovations, such as RankBrain, E-A-T, BERT, and MUM, directly or indirectly contribute to this goal. Google's efforts to develop a semantic search engine can be traced back to 1999. With the introduction of the Knowledge Graph in 2012 and the fundamental changes in the ranking algorithm (commonly known as Hummingbird) in 2013, it became more specific. With the introduction of the "Semantic Experiences" program, the company began to demonstrate its semantic capabilities. But Google is not the only company doing this. Microsoft's AI For Scale program also heavily relies on semantic search. The software giant writes in its company blog: We refer to this transformational capability as semantic search, which is the primary way we provide AI at scale to our customers. Although Stack Overflow recently integrated semantic models into its search, some people predicted last year, when ChatGPT gained popularity on the internet, that combining semantic search with language models could be a suitable way to achieve a better search experience. When semantic search and generative AI work together, they can improve accuracy, credibility, and the convenience of staying up-to-date with research. Companies that lag behind in adaptation or choose to adopt generative AI without incorporating semantic search and other technologies will struggle in fierce competition. To gain an advantage in competition, one must be at the forefront, and leveraging the combined power of semantics and language models can provide strategic advantages. Source: https://analyticsindiamag.com/what-semantic-search-can-do-for-llms/

RECENT AI NEWS

RECENT AI TOOLS