We have noticed within our circle of colleagues that during a typical workday, the main part of our tasks revolves around problem-solving.Questions like: Where can I find the appropriate presentation for a specific client meeting? Where can I find the relevant reference story? Which candidate should we hire? These questions keep recurring. And we have to admit that we will encounter them again in the coming months, albeit with slight variations in context and the technologies involved.Imagine how great it would be if we could simply ask an application, utilizing GPT or a similar language model as its foundation: How did I solve this problem last time? Or even better: How did the last person in our company solve this problem? If this sounds like a distant future to you, you are mistaken. There are already many domain-specific systems. The current advances in Artificial Intelligence (AI) and specifically in Natural Language Processing (NLP) make it possible to further develop these systems to provide solutions for general information retrieval.In this article, we will explore real-life case examples of recommendation and information retrieval systems in companies. Our goal is to demonstrate that the tools and technologies are already mature, and these systems can be easily reused in a variety of new use cases. Additionally, we will provide practical suggestions on how to overcome existing organizational barriers and successfully introduce these technologies in your company.
Examples of existing systems for recommendation and information retrieval: Content management platforms for customer service and sales.
Large companies are already implementing various recommendation and information retrieval systems to effectively utilize their accumulated knowledge. At daenet, we have been engaged in leveraging language models such as Microsoft Azure OpenAI’s ChatGPT in a secure and compliant manner to make it accessible for customers, enabling quantum leaps in information processing.
Content Management Platforms
Companies can leverage the offerings of Microsoft OpenAI to search for specific information such as customer materials and presentations through content management platforms. OpenAI provides a wide range of tools and APIs that can assist companies in finding specific information not only through textual search but also through semantic relevance.
Microsoft OpenAI also offers tools and APIs for information retrieval, which can help companies locate specific information. Semantic similarities between content can be utilized. Companies can utilize these offerings to search for relevant information within their internal databases, knowledge bases, or external sources through content management platforms and correlate them. This enables targeted searching for customer materials, presentations, or other relevant content.
Analysis and Categorization
Companies can use OpenAI to analyze and automatically categorize texts via content management platforms. This makes it easier to identify and retrieve relevant information such as customer materials or presentations. Even more interesting is the quick access to novel decision-making foundations, as all relevant data is transparent and available within a short period, even when searching for complex contextual relationships.
By enabling the analysis and categorization of data and content, we can create a matrix clustering with thematic focuses. This, in turn, becomes a highly effective tool for strategic corporate management.
Companies can integrate language models like OpenAI’s ChatGPT into their own chatbot systems to effectively respond to customer inquiries based on content management content. By utilizing advanced language models, the chatbot can access a broad knowledge base and deliver accurate and relevant information regarding content management platforms. This improves customer service and enables companies to provide fast and precise answers to customer questions.
It is important to note that the use of services like OpenAI’s language models for information search should be accompanied by proper data management and privacy measures to ensure that sensitive information is appropriately protected.
Here are some common types of searches:
- Keyword Search: Users can enter keywords or phrases to search for relevant content. The search engine then scans existing knowledge bases, ticket histories, and other resources for matches.
- Full-Text Search: This type of search scans the entire text content of documents, tickets, or knowledge bases for matches with the entered search terms. Complex queries can be used to find specific information.
- Filter-Based Search: Users can filter search results based on certain criteria or properties. This can be done, for example, by date, category, priority, or other metadata to increase the relevance of the results.
- Auto-Completion: While typing, the system provides possible suggestions or completions to facilitate the search. This is based on frequent search queries or known keywords.
- Category or Tag Search: Information can be organized into specific categories or tagged. Users can search within these categories to find relevant information on a specific topic.
- Semantic Search: The content of documents is found based on semantic similarity to a search query or example documents. This allows for discovering connections that a full-text search may not uncover.
The exact nature of search in customer-specific information systems can vary from organization to organization and depends on the tools and platforms used.
The application of similarity search is not limited to enterprise-specific systems. Many incident management solutions offer such features. An example of this is Azure Sentinel, a software for managing security incidents. Microsoft recently released the “Similar Incidents” feature in preview. This provides administrators with a list of previously resolved tickets related to similar issues, which can provide helpful information when handling current incidents. This integration of similarity search facilitates the efficient handling and resolution of recurring or similar problems in incident management.
Experiences through Information Retrieval
The field of Information Retrieval (IR), which has been intensively researched for around 50 years, enables effective searching of unstructured documents, primarily in text form, to meet information needs within large collections. A well-known example of an IR system is Google, which delivers excellent results for web searches. However, these systems are not yet equally suited for enterprise-related, institutional, and domain-specific search requirements. Only in recent years has the importance of this scenario been increasing.
Thanks to the progress of large-scale language models, with ChatGPT being the most well-known example, searching complex enterprise data stores is becoming easier. Additionally, the big data ecosystem is already highly mature, allowing infrastructure and data engineering challenges to be addressed with standard tools. These factors have led to the topic, which was originally relevant only to early adopters, now gradually being embraced by the majority of companies, and with the right implementation partner, it is no longer science fiction but can provide a real competitive advantage.
Practical Applications of these Tools
Recent advancements in artificial intelligence greatly simplify the application of information retrieval techniques. Modern models already possess the semantic understanding that previously required extensive manual feature engineering. If you are not yet using these techniques in your daily work, you should seriously consider doing so.
Most industries face the challenge of recurring problems that occur multiple times within a few months. An intelligent system could handle most tasks related to problem-solving with minimal human supervision. This only requires access to digital records describing how similar problems were solved in the past.
There are a variety of existing technological foundations to build upon
ElasticSearch “You may also like” feature: This query method in the popular ElasticSearch platform allows for finding documents similar to a given set of documents.
Azure Cognitive Search: A search engine by Microsoft, similar to ElasticSearch but with many options for AI-based extensions and integrations with various other Azure services.
OpenAI Embeddings and Search: These are APIs from OpenAI that are based on neural models and enable the semantic comparison between a search query (what you are looking for) and a set of documents (where you are searching).
Open Capivara: An open-source project that has developed a cognitive layer for searching support tickets.
Reasons for the hesitant adoption of information retrieval systems
Despite the availability of various tools, you may wonder why not every company is using them. There are different perspectives that explain why organizations struggle with the adoption:
From the perspective of engineers and developers: The architecture of information retrieval systems differs fundamentally from most other software solutions and requires specific expertise for implementation. The evaluation and testing methods also pose challenges in verifying the correctness of the developed system.
From the perspective of users: Most people are not accustomed to thinking about “information retrieval from large databases” in their work. Therefore, they may have difficulty fully realizing the benefits of IR systems. If the user interface is not intuitive or the system’s behavior is unclear to users, even the best IR systems will not be widely accepted.
From the perspective of managers and product owners: Integrating IR functionality into an existing system requires not only technical knowledge to develop the features but also a strategic vision of how these features can create value and how their development should be prioritized. Managers and product owners often lack this vision and prefer more traditional technologies.
The good news is that the current surge of innovation in artificial intelligence has significantly lowered the technical barriers to adopting highly powerful IR systems. This makes it possible, for example, through prototyping or concrete problem-solving examples, to make the value of modern IR systems more tangible and experiential for users and managers.
Therefore, we advise companies to dive into this new world with a suitable technology partner to unlock competitive advantages.