↑↓ selectEnter openEsc close

    1. Elbuz

    How specifications from a website become attributes in your catalog

    13 min readUpdated

    Elbuz moves parser-collected specifications into your attribute directory: known names land in existing attributes, and the mapping dictionary and “AI match” keep out duplicates.

    When you need this:

    • Sites name one specification differently: one rule in “Attribute mapping” maps them to your attribute for all sources.
    • Hundreds of new names: “AI match” applies confident pairs itself and asks you to confirm the rest.
    • A site value belongs in a card field, such as “Manufacturer”, not an attribute.
    • The parser created extra attributes: “Ignore” plus deletion in the right order removes them for good.

    Where to start: after the first collection, open “Attribute mapping” and sort out “List of new attributes”.

    What specification mapping does

    How specifications get into the catalog

    What's collected stays in the source price list: the itself moves nothing into your catalog. That's done by the base catalog update, a separate run with its own settings.

    The update starts by itself if the “Update catalog after import” setting is on (it's on by default): after all price lists are imported on schedule, and if is still running, a minute after it ends. So deletion marks left in the update window “just to check once” will fire at night without you: five ways the update starts.

    The parser looks for specification rows inside the block from “Selector #1”, so site menu items don't end up in specifications. If “Selector #1” is empty or no rows are found in the block, the search covers the whole page except the header, menu, and footer (the header, nav, and footer tags).

    A row where the name and value come as one text (“Color: black”) is split at the first colon. Such a row without a colon is dropped: it's a heading or a phrase, not a specification.

    Specifications become catalog attributes

    Each collected specification name looks for an in your attribute directory. The order is:

    1. names from the mapping dictionary marked “Ignore” are deleted;
    2. names that have a pair in the dictionary get your attribute, and the specification name itself is replaced with the catalog attribute name;
    3. next, a match by attribute name, with the main synonym taking priority;
    4. next, a match by the checksum of the name;
    5. everything that wasn't matched is created in the directory as a new attribute. If the “Attribute block name” field was collected for the specification, a block with that name is also created in the directory and the attribute goes into it; when the block name is empty, it goes into the default block.
    Checks run from top to bottom, and a name stops at the first one that works. Everything none of them recognized reaches the bottom “New attribute” box
    yesnoyesnoyesnoyesnoSpecification namein the source price list“Ignore” is setin “Attribute mapping”The row is deleteddoesn't go into the catalogThere is a pair in the mappingdictionaryAttribute from the pairname becomes the catalog attributenameThe directory has an attributewith this nameExisting catalog attributemain synonym firstMatch by name checksumNew attribute in the directory
    • Start or end
    • Action or result
    • Decision: the path depends on the answer

    The last step is why the attribute directory of those who parse grows by itself: sites call the same specification different names, and each variant becomes a separate attribute.

    Moving photos and specifications from a parser source into the catalog costs money: €0.05 per product, once. Regular file price lists are free. The “Load attributes” checkbox hint warns: “An extra charge applies when updating the catalog”. The amount before the run and a short balance: How to update your catalog from supplier price lists.

    How to map site names to your attributes

    You do this in the “Attribute mapping” window.

    This is the mapping dictionary. It has two tabs: “List of new attributes” and “List of matched attributes”. The hint above the grid tells you what to do: “To map site attributes to base catalog attributes, double-click the Base catalog attribute name field and select the attribute from the drop-down list.”

    Attribute mapping window with the tabs for new and matched attributes, the hint and the counters strip
    1Ignore column
    2Attribute name from site (source)
    3Base catalog attribute name
    4AI match button
    1. Ignore — tick to skip a site attribute.
    2. Attribute name from site (source) — the name as the supplier site spells it.
    3. Base catalog attribute name — the catalog attribute it is matched to.
    4. AI match — proposes matches for new names.

    Columns:

    • “Ignore”: the specification doesn't go into the catalog at all. Put here things that aren't really specifications, such as “Product on brand website” or “Official website” rows, which sites also put in the specifications table.
    • “Attribute name from site (source)”: what it's called on the other site.
    • “Base catalog attribute name”: what it's replaced with in your catalog.
    • “Category”: the rule applies only to this catalog category.
    • “Base catalog field name”: which card field (for example, “Manufacturer”) gets the value instead of an attribute. Works only in content search for existing products (details); it doesn't affect moving the price list into the catalog.

    The dictionary is shared by all sources: a rule created on one parser works on all of them. That's by design: suppliers write “Diagonal”, “display diagonal”, and “screen size”, but you have one attribute.

    An attribute from the site straight into a card field, for example “Manufacturer”

    A site specification can go into a card field instead of an attribute: “Manufacturer”, “Product name”, short or full description, meta tags, . Use the “Base catalog field name” column in “Attribute mapping”: in the site attribute row, click the arrow right of the cell and choose a field.

    Attribute mapping window with the list of product card fields open from the Base catalog field name column
    1Base catalog field name column
    2Manufacturer field in the list
    1. Base catalog field name — the column whose arrow opens the list of product card fields.
    2. The list of card fields, filtered by the word Manufacturer; picking an item saves the mapping at once.

    The rule works only in content search for existing products: in the “Product content search” window, the “Fill catalog fields from attribute mapping” checkbox must be selected in the “What to save” block. Then on “Save” the attribute value goes into the chosen card field.

    There's a condition for “Manufacturer”: the value from the site must match the name of a manufacturer that's already in your manufacturer directory (case doesn't matter).

    The app doesn't create a new manufacturer: if there's no match, the field doesn't change and the product keeps its previous manufacturer. With “Do not overwrite filled fields” on, only empty fields are filled. The column doesn't affect moving a parser source price list into the catalog.

    Where to see all links and delete a wrong one

    All mapped site attributes are on the “List of matched attributes” tab of the same “Attribute mapping” window. The same list opens as a separate “List of matched attributes” window from the attribute directory: Attributes → “AI tools” → “Find duplicate attributes using AI” → “List of links”. The button shows when the selected category has attributes.

    The List of Mapped Attributes window: a toolbar with a Delete icon and a grid of synonym attributes, their main attributes and an Active flag.
    1Delete icon
    2Attribute synonym column
    3Main attribute column
    4Active column
    1. Delete icon — removes the selected links; do not use it on the shared dictionary.
    2. Attribute synonym — the name as it appears on the supplier site.
    3. Main attribute — the catalog attribute the synonym is linked to.
    4. Active — whether the link is applied during import and parsing.

    The window has three columns: “Attribute synonym” is the name from the site, “Main attribute” is your attribute it's replaced with, and “Active”. You can filter rows by name in the filter row above the column.

    • Delete a wrong link: select the row, click the trash icon (“Delete”) in the toolbar, and confirm. On the next run, the site name becomes a new specification again, and you can map it correctly.
    • Turn off temporarily: clear the “Active” checkbox. An inactive link isn't applied during mapping.

    How to map hundreds of names with the “AI match” button

    Sorting out several hundred new names by hand takes a long time, so the toolbar has AI matching. First the app builds an index of the attribute directory (until it exists, the window shows “Indexing attribute dictionary…”), then it picks candidates for each name and asks the model whether it's the same specification.

    The answer gets a confidence score:

    • 92 and higher: the mapping is applied automatically and marked “Auto-matched” in the grid;
    • 75 to 91: this is a suggestion. Two buttons appear in the row, “Confirm match” and “Reject (new attribute)”. To confirm all suggestions at once, use the “Confirm all” button;
    • below 75: the specification is considered new.

    AI never overwrites a manual mapping. The model isn't allowed to merge opposite sides of one parameter: “Width”, “Height”, and “Depth”, “Net weight” and “Gross weight”, “Min. temperature” and “Max. temperature”, “Inner diameter” and “Outer diameter” are different attributes. Five counters are shown above the grid: “Auto-matched”, “Suggestions”, “New”, “Manual”, “Unmatched”.

    What the line under the button shows

    After you click “AI match”, a status line appears under the button, and the button stays inactive until the job is done:

    • “Starting AI matching…” — the job is being queued;
    • “AI attribute matching in progress…” and “Names processed: N of M” with a progress bar; after about 20 seconds, “≈ N min left” appears next to it;
    • “Done: N names processed. See the table and counters below.” — the summary stays under the button while the window is open.
    Attribute matching window: the AI match button and the status line saying there are no new names to match
    1AI match button
    2Status line
    1. AI match — the button was pressed once; it starts matching of new attribute names.
    2. Status line — reports that all names are already in the matching dictionary, so there is nothing new to match.

    If there's nothing to do, the line gives the reason: “No new names to match: all names are already in the matching dictionary.” or “AI is not available for your account — matching was not started.”

    A manual run doesn't wait for the parser to finish collecting; it matches the names already collected. Only the automatic matching after a run waits for collection to end.

    Why “Product content search” won't let you save while matching is running

    Content search maps found specifications to the directory itself, right after the run. It's turned on by the “Map specifications to the catalog” checkbox: the “Run” tab → the “How to run” step → the “AI check” group. It's on from the start.

    If you clear it, a site name with neither a pair in the mapping dictionary nor a same-named attribute becomes a separate attribute after “Save”, even if you have that specification under another name. Exception: with “AI attribute matching” on in the general settings, matching runs without the checkbox.

    For matching, the app keeps an index of the directory. New and renamed attributes are added to it automatically: every night at 03:40 and after each run with matching, if the index already exists (up to 300 names per run).

    The nightly build-up takes about 8 minutes; the rest carries over to the next night. It runs only if matching is on for at least one user and the content search “Results” tab has found specifications.

    Google search needs the same index: the summary table immediately gets the category's specification names and directory names close in meaning (how the table is built).

    While it works, the “Results” tab shows “Matching found attributes against the dictionary” with a counter, and if the directory isn't ready yet, “Preparing the attribute dictionary for matching…”. The same lines are written to the log on the “Progress” tab.

    During matching itself (“Matching found attributes against the dictionary”), “Save” is inactive; clicking it shows: “Attribute matching is still running. Wait for it to finish: otherwise some attributes will be created in the catalog a second time, as separate entries.” Not an error: the app won't write specifications to cards before deciding which are existing attributes and which are new.

    At the earlier stage, “Preparing the attribute dictionary for matching…”, the button is still active, but save after matching for the same reason. The wait depends on the number of new names: watch the counter. The button becomes active again when:

    • matching has finished: “Attributes matched against the dictionary”;
    • the window has handed the rest over to the background queue: the log shows “Attributes left to match: N. They are being processed by the background queue.”;
    • the next step has failed: the log shows “Attribute matching is not progressing…” (see below).

    The summary sits in the filter row above the product list: the “New attributes” link with a counter. Its hint shows how many were “Matched automatically by AI” and how many are “Awaiting your confirmation”; clicking it opens the confirmation window with the buttons described above.

    If instead of the counter you see “Attribute matching is not progressing: the job queue is not responding. Nothing is lost — it will continue once the queue is running.” (in the “Attribute mapping” window: “Matching is not progressing: the job queue is not responding…”), the server didn't respond to the next step.

    What's done is saved in the directory, and the background queue picks up the remaining names. You can close the window and come back later: open “New attributes” and check whether “Unmatched” has gone down. Clicking “AI match” again is safe: it takes only names that aren't in the directory yet and doesn't overwrite what's already mapped.

    Until the previous job finishes, clicking “AI match” again doesn't create a new one. In the “Job queue” window, this job is shown as “AI attribute matching”; “AI product matching” there is a different job, matching supplier products to the catalog.

    How to remove attributes and groups created by the parser

    An attribute created from parser data has no “came from the parser” mark: it looks like one created by hand. Delete it the same way, with the trash icon in the attribute directory, with the same consequences. Gone with it: its values in all products, category links, catalog grid column, formulas using it, export template field. There's no undo.

    What's special about the parser is that a deleted attribute comes back. The catalog update looks for an attribute by specification name, and a name that no longer matches anything is created as a new attribute. So remove it in this order:

    1. in the “Attribute mapping” window, set “Ignore” on that name or map it to the right catalog attribute;
    2. only then delete the attribute from the directory.

    Each step alone won't work:

    • “Ignore” by itself removes nothing from products. The mark only drops the name from price list data during the catalog update, and the values products already have stay until you delete the attribute;
    • if you delete the attribute without “Ignore”, the next catalog update after a new price list collection creates that name as an attribute again;
    • if the name was mapped to exactly the attribute you deleted, that mapping stops working, and the name is also created as a new attribute. You need to map it again.

    A group created from the “Attribute block name” field is deleted with the trash icon on the “Attribute groups” tab. The group's attributes aren't deleted: they move to the group marked “Default”.

    The group is also removed on the connected site if “Delete data immediately on the site” and “Attributes” in the “Select data to delete on the site” block are on in the site export template. The “Default” group itself can't be deleted.

    But the catalog update creates a group by name just like an attribute. As long as the parser collects this block name, the group will reappear after the next price list collection and catalog update. To prevent this, clear the selector of the “Attribute block name” field in the source settings.

    Scenarios

    Fill the catalog with specifications without bloating the attribute directory

    What to do. In the “Attributes” group, fill in the “Operation selector” row (specifications block in “Selector #1”, one row in “Selector #2”), plus “Attribute name” and “Attribute value”. After collection, before the catalog update, open “Attribute mapping”: on “List of new attributes”, map the names you need to catalog attributes and mark extra ones “Ignore”.

    What you get. The base catalog update writes the values into your attributes: a mapped name is replaced with the catalog attribute name, and a name marked “Ignore” doesn't go into the catalog.

    What won't happen. Without “Selector #2”, specifications aren't collected; the only exceptions are <dl> markup and a JSON source (about specifications). A name that has neither a pair nor an attribute with the same name is created as a new attribute.

    Write the brand from a site specification into the “Manufacturer” field

    What to do. In the “Attribute mapping” window, in the site attribute row, click the arrow in the “Base catalog field name” column and choose “Manufacturer”. In the “Product content search” window, in the “What to save” block, turn on “Fill catalog fields from attribute mapping”.

    What you get. On “Save” in content search, the attribute value goes into the card's “Manufacturer” field if a manufacturer with the same name (regardless of case) is already in your manufacturer directory.

    What won't happen. The app doesn't create a new manufacturer: without a match, the card keeps its previous manufacturer. This rule doesn't apply when a parser source price list is moved into the catalog.

    What to do when the attribute directory has grown or matching is stuck

    What you see in the directory and what to do about it
    What you seeWhyWhat to do
    After a catalog update, dozens of new rows have appeared in the attribute directory A specification name without a pair is created as a new attribute Sort out the “List of new attributes” before the next update: map the ones you need, mark the extra ones “Ignore” (how to map names)
    “Attribute matching is still running. Wait for it to finish…” Content search is still mapping the specifications it found to the directory; saving now would create them as second attributes Wait until the “Matching found attributes against the dictionary” line disappears, then click “Save” again (about matching in content search)
    “Matching is not progressing: the job queue is not responding. Nothing is lost…” The server didn't respond to the next AI matching step. What's done is saved; the background queue picks up the rest Don't restart anything right away: come back later and check the “Unmatched” counter. Clicking “AI match” again takes only unmatched names (details)
    You clicked “AI match”, but matching didn't start The line under the button gives the reason: there are no new names — all collected names are already in the matching dictionary; or AI isn't available for the account No new names — nothing to do: new ones will appear after the next collection. AI isn't available — contact support (what the line under the button shows)
    A deleted attribute appeared again after a catalog update The parser still collects this name, and the catalog update creates a name without a pair as a new attribute First mark the name “Ignore” or map it, then delete the attribute (how to remove attributes)

    FAQ

    Specifications were collected, but the product doesn't have them in the catalog. Why?

    What's collected moves into the catalog with the base catalog update. Check whether it ran, whether these names are marked “Ignore”, and whether your balance covered the transfer (details).

    Can I trust the “AI match” button?

    A mapping with confidence 92 or higher is applied automatically, 75 to 91 waits for your confirmation, and below 75 the name is considered new. AI doesn't overwrite a mapping made by hand.

    Does a mapping created on one parser work for others?

    Yes. The mapping dictionary is shared by all sources: a rule created on one parser works on all of them.

    Does “Ignore” remove values that products already have?

    No. The mark only drops the name from price list data during the catalog update. To remove the values from products, delete the attribute itself after “Ignore” (order of steps).

    Do I need to turn on AI matching in content search?

    No, the “Map specifications to the catalog” checkbox is on from the start, and the directory index is built up automatically every night. Clear it only if you want every new name to become a separate attribute (details).

    Can I turn off a link without deleting it?

    Yes. In the “List of matched attributes” window, clear the “Active” checkbox: an inactive link isn't applied during mapping.

    Where to find it and who has access

    Menu
    Directories → Processing price lists → List of price lists
    Attribute mapping
    the source row → the arrow next to “Product matching” → “Attribute mapping”
    From the source card
    the “Products” tab → the additional operations menu → “Attribute mapping”
    List of links
    Attributes → “AI tools” → “Find duplicate attributes using AI” → “List of links”
    Access right
    the “Price lists” window; there's no separate right for attribute mapping
    Prepare
    a parser source with collected specifications

    How to set up collecting specifications on a site: see What exactly the parser takes from a product page.

    Related topics

    Was this article helpful?

    Missing a feature? Tell us.

    We will add it for free if other users need it too.

    If only you need it, we will agree on it separately.

    Article contents

    Suggest an improvement

    Try Elbuz free