Stop testing, start deploying your AI apps. See how with MIT Technology Review’s latest research.
Download now
We now have the ability to arbitrarily search for words in our indexed documents. But searching is only the first step in retrieving information that we’re looking for. After we have a list of documents, we need to decide what’s important enough about each of the documents to determine its position relative to other matching documents. This question is generally known as relevance in the search world, and one way of determining whether one article is more relevant than another is which article has been updated more recently. Let’s see how we could include this as part of our search results.
If you remember from chapter 3, the Redis SORT call allows us to sort the contents of a LIST or SET, possibly referencing external data. For each article in Fake Garage Startup’s knowledge base, we’ll also include a HASH that stores information about the article. The information we’ll store about the article includes the title, the creation timestamp, the timestamp for when the article was last updated, and the document’s ID. An example document appears in figure 7.4.
With documents stored in this format, we can then use the SORT command to sort by one of a few different attributes. We’ve been giving our result SETs expiration times as a way of cleaning them out shortly after we’ve finished using them. But for our final SORTed result, we could keep that result around longer, while at the same time allowing for the ability to re-sort, and even paginate over the results without having to perform the search again. Our function for integrating this kind of caching and re-sorting can be seen in the following listing.
def search_and_sort(conn, query, id=None, ttl=300, sort="-updated",
start=0, num=20):
We’ll optionally take a previous result ID, a way to sort the results, and options for paginating over the results.
desc = sort.startswith('-')
sort = sort.lstrip('-')
by = "kb:doc:*->" + sort
Determine which attribute to sort by and whether to sort ascending or descending.
alpha = sort not in ('updated', 'id', 'created')
We need to tell Redis whether we’re sorting by a number or alphabetically.
if id and not conn.expire(id, ttl):
id = None
If there was a previous result, try to update its expiration time if it still exists.
if not id:
id = parse_and_search(conn, query, ttl=ttl)
Perform the search if we didn’t have a past search ID, or if our results expired.
pipeline = conn.pipeline(True)
pipeline.scard('idx:' + id)
Fetch the total number of results.
pipeline.sort('idx:' + id, by=by, alpha=alpha,
desc=desc, start=start, num=num)
Sort the result list by the proper column and fetch only those results we want.
results = pipeline.execute()
return results[0], results[1], id
Return the number of items in the results, the results we wanted, and the ID of the results so that we can fetch them again later.
When searching and sorting, we can paginate over results by updating the start and num arguments; alter the sorting attribute (and order) with the sort argument; cache the results for longer or shorter with the ttl argument; and reference previous search results (to save time) with the id argument.
Though these functions won’t let us create a search engine to compete with Google, this problem and solution are what brought me to use Redis in the first place. Limitations on SORT lead to using ZSETs to support more intricate forms of document sorting, including combining scores for a composite sort order.