The Partium Search Engine 2022 introduces our new, state-of-the-art, proprietary, and patented machine learning engine with our novel semantic image search, semantic text search, and exact text search capabilities
What's new?
Semantic image and text search
Semantic image and text search will level up the part search for Partium users by improving usability, enhancing performance, and minimizing implementation effort.
With the new Search Engine, we're advancing the search capabilities on two major points:
- Semantic Image Search: The new engine finds parts without having reference images in your master data. This enables companies to unlock Partium's full potential without investing resources to capture reference images.
- Semantic Text Search: Previously, the Text Search was finding parts by comparing the text query with the technical descriptions in your master data. The new Search Engine finds parts that are semantically related to the text query by translating a textual description into concepts.
This enables the user to search for synonyms or visual descriptions of reference images, and thereby supports users that do not know the exact technical terms of the part to execute successful searches.
Exact text search filter (search for IDs or measurements)
In addition to the semantic search, we're adding an exact search filter capability to find parts using IDs, measurements, manufacturer brand names, or any other attribute available in the master data. By using this search filter, the engine will exclusively show results where the master data matches the Exact Text query. If your technicians are used to finding parts with IDs or fragments of IDs, or if the part's size, voltage, or power is known, the exact search capability will filter results based on these specific text queries.
How do you use the Partium Search Engine 2022?
After login, the web app shows three input fields as in the screenshot below.
1. Search with a description box: used to trigger a Semantic Text Search.
2. Search with exact terms box: used to trigger an Exact Text Search.
3. Drag and drop image to search with image or browse files box: used to trigger a Semantic Image Search.
Once a search is initiated, a result list is populated. By adding further search inputs, such as additional text terms, tags (4), or filters (5), the result list adjusts dynamically. The tags are generated based on the semantic text search query and when clicking on one, it is added to the search with a description box.
Search Examples
Below are some examples of different search queries, exploiting multiple combinations of the search modalities available in the new engine.
1. Search with description (semantic text search)
The query “ventilator” will also show fans in the result list because they are conceptually the same kind of part.
The query “red handle” will show semantically similar parts.
2. Semantic image search without reference image + description
The image query + a short description finds the part although no reference image is available (Part PM1 has only a thumbnail picture but no reference images).
3. Semantic image search + semantic text search
With the image query plus the semantic text query “zinc”, the engine finds parts that are a combination of both.
4. Search with measurements and IDs
With the exact search box, the user can perform advanced searches for measurements or IDs. Here are some examples to showcase some of the possibilities.
- Query: “15 mm” à results will show parts that contain “15 millimeters”
- Query: “8 Volts” à results will show parts that contain “8V”
- Query: “230A” à results will show parts that contain “230A220V”
- Query: “9’ ” à results will show parts that contain “9 in”
- Query: “in” à results will show parts that contain “inches”
- Query: “02213” à results will show parts where the query is part of the ID
How does the Partium Search Engine 2022 work?
The latest advances in Partium's state-of-the-art, proprietary, and patented machine learning engine enables the Partium Search Engine 2022 to find spare parts using our novel semantic image and text search. The machine learning engine derives concepts from images and/or text descriptions from master data or search queries to find parts more easily, accurately, and flexibly. As a result, the engine is capable to relate a text query to reference images in the master data or relating image queries to technical descriptions in the master data. Additionally, the engine can combine image and text queries to achieve the best results and to provide a more accurate ranking of the possible matches.
- Semantic Image Search: The engine will search parts only with the convenience of an image that has a lot of information in a single shot. It will then suggest results where the image query is semantically related to textual information or/and available images in the master data. The Semantic Image Search is best suited for standard parts for which there is an abundance of data that Partium can leverage in their machine learning models. Consequently, for standard parts, the engine can cope with no reference images in most cases. This is very convenient for companies that do not have reference images for their spare parts available. When it comes to custom parts, the need for reference images is more relevant because those parts can only be found using our Visual Matching engine.
The Visual Matching Engine works with visual similarity between the query image and the reference image, thereby taking the visual appearance of the object itself into account. The Partium Search Engine 2022 seamlessly couples Visual Matching and Semantic Image Search to produce the best results for users with both technologies enabled. Note that Visual Matching requires reference images. - Semantic Text Search: The engine will search parts that are semantically related to the search query. Instead of trying to find the exact search term within the master data, the engine relates the semantic concept present in the combination of terms in the text query with the concept encompassed by the technical description and reference images (if any) in the master data. This enables the Partium Search Engine to find parts with search terms that may or may not coincide with the part names or any other portion of master data imported into the Partium systems. This is very convenient when the user does not necessarily know the exact part descriptions given in the master data (e.g. new employee, limited language skills, ...) or for incomplete / non-consolidated master data (e.g. due to a company merge, mixed terminology, etc.)
In addition to the semantic search capabilities, the engine contains an Exact Text Search to filter the results where the query words are present in the master data Examples are technical terms such as IDs, manufacturer brand names, model numbers, or quantities. Of special importance is our new handling of measurements. This is very convenient when the user knows that information with high certainty.