Search Functionality: Beyond Meals

Advanced search functionality stands at the core of customer satisfaction! If you don't trust me look at mayor search engines... Food delivery platform are now embracing a far broader scope, integrating new verticals such as groceries, pet supplies, alcohol, flowers and more into their offerings. This shift from a food-restaurant-model to a marketplace requires an overhauling the search functionality.

The journey from simple, nested food menus to stock keeping units (SKUs) marks a significant leap in the complexity of online search systems. This transition underscores the necessity of developing a detailed data layer capable of indexing new verticals efficiently. As the inventory diversifies, the challenge of accurately understanding and interpreting customer search intent escalates. Whether a customer searches for "chicken" - seeking raw ingredients for a recipe or a ready-to-eat chicken meal, the search engine's ability to discern and respond to such intents becomes crucial for delivering relevant results and enhancing customer experience.

Search Infrastructure for Scale and Diversity

Data catalogs play a vital role in highlighting valuable assets, merely cataloging these assets falls short of meeting the real needs of customers. After all, customers search for "things" and insights, not just mere "strings" of text.

This search process is heavily influenced by an individual's mental model of their problem, shaping their search terms and overall behaviour. To genuinely decode their intent, data catalogs must transcend traditional keyword matching. They should act as bridges, connecting the customers's conceptual understanding of their problem with the domain.

1. Constructing Ontologies: Developing structured representations of knowledge, or ontologies, instead of settling for basic glossaries. This approach enables a deeper understanding of how different data elements interrelate, enhancing the comprehensiveness of search results.
2. Implementing Semantic Search: Utilising semantic search techniques that understand the intent behind queries, rather than solely focusing on literal keywords. This leads to search outcomes that are significantly more aligned with the customer's actual needs.
3. Introducing Conversational Interfaces: Adopting conversational search interfaces that interact with customers more naturally.

Machine Learning-Based Approaches

To tackle these challenges, food delivery platforms are turning to machine learning-based approaches. Learning-to-rank systems and query/document understanding enhancements are being integrated to refine search relevance. These technologies not only facilitate a deeper comprehension of customers queries but also enable the personalisation of search results, making them more pertinent to individual customers' needs and preferences.

An ontology—a structured framework of categories and relationships—enables the search engine to grasp the nuanced meanings and contexts of customers queries. By systematically categorising dishes, restaurants, food and menus within a well-defined hierarchy, an ontology helps in accurately interpreting customers intent, even when queries are ambiguous or complex.

This structured understanding supports the machine learning models in drawing more meaningful inferences about customers preferences, thereby significantly improving the accuracy and relevance of search results. For instance, when a customer searches for "vegan options," the ontology-based system can intelligently filter and prioritise results that align with vegan dietary preferences, leveraging its understanding of the relationship between "vegan" as a dietary category and relevant menu dishes or restaurants.

The integration of an ontology with machine learning-based approaches such as learning-to-rank systems and query/document understanding enhancements offers a more dynamic and context-aware search experience.

Rebuilding Search Infrastructure

Addressing the need for scalability, the search infrastructure itself has undergone significant reconstruction. Next-gen indexing techniques shorten the time required to refresh search indexes, ensuring that customers have access to the most current offerings. Federated search systems allows for the efficient retrieval of results across multiple verticals, from restaurants to groceries, streamlining the customers experience by providing results in a single query.

Personalising Search Results

The personalisation of search results with customers data such as profiles, previous searches, orders, location, entry page, page visits and history can help determine better search results. By assigning different weights to these features, developers can tailor search outcomes to better match individual customer behaviour and preferences.

For example: If a customers was visiting a Restaurant that sells Fried Chicken, and then searches for chicken we can assume that we should prioritise results for dishes in restaurants on the other hand if a customers entered through an Ad from Google to a specific grocery landing page and then navigates to the main homepage and searches for Chicken; we can use the Google Adwords campaign, entry page to understand that he might be looking for a grocery product. If a customer has ordered several times vegetarian burgers and he searches for burgers, we might want to customise the search to display vegetarian burger.

We can take off some of the heaving lifting by using Amazon Personalize plugin with OpenSearch as discussed in the blog:

https://aws.amazon.com/blogs/machine-learning/personalize-your-search-results-with-amazon-personalize-and-amazon-opensearch-service-integration/

Advancing Query and Document Understanding

Expanding beyond restaurants & dishes, the adaptation and extension of data labels and taxonomies to new domains are essential (building an early-ontology model). Leveraging machine learning labelling services and human annotations, systems can enrich their "understanding" of diverse product ranges.

https://aws.amazon.com/sagemaker/groundtruth/

https://www.mturk.com/

Amazon Mechanical Turk (MTurk) is a crowdsourcing marketplace that makes it easier for individuals and businesses to outsource their processes and jobs to a distributed workforce who can perform these tasks virtually. This could include anything from conducting simple data validation to a turnkey data labeling service that provides an expert workforce and manages it on your behalf.

Beyond improving recall, refining the ranking process is crucial for presenting customers with results that align with their needs. Traditional ranking based on string similarity has shown to be inadequate for non-branded queries, requiring the development of dedicated ranking models that leverages machine learning techniques to consider multiple aspects of search context. By treating queries as "things" - objects in an ontology- rather than "strings" - keyword- platform can significantly improve search relevance.

Vector Searches and Embeddings

Vector searches and embeddings represent a significant advancement in search technology, enhancing accuracy by understanding the context of search terms. This approach transforms textual queries into high-dimensional spatial searches, enabling the search engine to pinpoint the most relevant results. Exploration of new search paradigms, including deep learning-based embeddings and approximate nearest neighbour searches, showcases the potential of these technologies to redefine search functionality.

More on modern search: https://aws.amazon.com/blogs/industries/building-blocks-for-modern-retail-ecommerce-and-media-search-with-aws/

Beyond Text: Incorporating Visual and Voice Search

As platforms evolve, so does the modalities through which customers interact with search functionalities. Incorporating visual and voice search capabilities represents the next frontier in creating intuitive and accessible search experiences. Visual search allows customers to search for products using images or barcodes, making it easier to find specific items without the need for precise textual descriptions. Similarly, voice search enables hands-free interactions, catering to customers needs for convenience and speed, especially when multitasking.

Integrating Customer Feedback for Continuous Improvement

The evolution of search functionality is not solely a technological endeavour but also a customers-centric process. Actively soliciting and integrating customer feedback into the development cycle is crucial for identifying pain points, uncovering hidden needs, and refining search algorithms. Platforms that establish effective channels for customer communication and feedback are better positioned to make iterative improvements that directly address customer preferences and challenges.

Continuous Improvement and Future Directions

The adoption of AI and the exploration of new search methods, such as visual and voice search, present new opportunities to transform search into a more engaging, conversational, convenient customer-focused activity.

The development and implementation of learning-to-rank algorithms, along with frameworks for evaluating the relevance of search results & personalisation, are crucial for advancing the performance.

This article highlights the need of optimising search methodologies to improve the overall customer experience. We invite readers to apply these insights to their own platforms, cultivating a culture of innovation. Whether sharing your own journey or seeking expert advice, engaging with the broader community can help tackle the challenges of advancing search functionalities, ensuring your service remains adaptable and attuned to the demands of your customers.

As we look to the future, the potential to revolutionise how customers discover and engage with online platforms is immense. By staying at the forefront of search technology advancements, platforms can ensure they remain competitive, relevant, and capable of delivering exceptional value to their customers. Let's continue to push the boundaries of what's possible, creating search experiences that are not just functional, but truly remarkable.