Search for content in message boards

Soundex is back bedeviling default NewSearch

Soundex is back bedeviling default NewSearch

Posted: 26 Jun 2012 2:32PM GMT
Classification: Query
Edited: 26 Jun 2012 2:41PM GMT
I thought this was taken care of -- by eliminating Soundex as part of the default.

My initial search from a tree was for surname Barrickman. The first 50 results included items (for wrong dates, considering this tree person died in 1804) for Bresnahan, Brezner, Briggen, Brogan, Brookner.

After the first 4 results, which could have been for the tree individual, the top 46 results were
--four impossible Civil War items (death in 1804) with 3 stars, but at least the name was right.
--All of the rest were 3-star items for City Directories; one for 1894, one 1895 and one 1896 (both right name but irrelevant places), the rest from the US City Directories, 1821-1989 (Beta) group.

Given the wrong dates and names, they should not have had any stars.

So: wrong for name, wrong for dates, wrong for places, and the correct 1790 and 1800 US Census results were missing from the top 50 results (right name, place, time period).

Here's the URL for the search results
http://tinyurl.com/6vt56po

Some change seems to have rolled back to the irrelevance of NewSearch in its first year or so of existence.

Is there a fix in the works?

All help appreciated,
Jade

Re: Soundex is back bedeviling default NewSearch

Posted: 26 Jun 2012 4:17PM GMT
Classification: Query
Hi Jade,

I'll investigate more to see if anything has changed (it shouldn't have). We replaced Soundex with our more restrictive phonetic matching algorithm. It is repsonsible for the variation in surnames you mention.

John

Re: Soundex is back bedeviling default NewSearch

Posted: 26 Jun 2012 4:25PM GMT
Classification: Query
Hi John,

Thanks for your reply.

A few of the results in the posted results-list would be somewhat reasonable phonetic simulations (which I could tolerate better as possible records and transcript variations), but not the rest on the list. The closer phonetic variations were also completely wrong as to time and place.

Best wishes,
Jade

Re: Soundex is back bedeviling default NewSearch

Posted: 27 Jun 2012 12:16PM GMT
Classification: Query
I have great admiration for Jade's persistence in continuing to use ranked search. As I have said many times, I abandoned ranked search many years ago because the ranking algorithm seems so completely irrational. So I really don't understand how anybody finds anything with ranked search.

The question of ranked search vs. exact search for the most part is not a New Search vs. and Old Search issue. Well, when New Search first came out it had such a strong bias towards ranked search that it made using exact search problematical. That situation eventually improved a great deal, plus there are now some really neat exact search features in New Search that were not in Old Search. But many of the time and place issues that Jade is describing bedevil exact search just as much as ranked search.

Jerry

Re: Soundex is back bedeviling default NewSearch

Posted: 27 Jun 2012 2:10PM GMT
Classification: Query
"persistence in continuing to use ranked search"--Jerry

Well, I was commencing a search for a different purpose regarding this individual.

Unfortunately, all ~initial~ searches from tree individuals are initially global and either ranked or by database, depending on how one's prior search was configured. It is not possible to define the search parameters before doing the initial search-from-tree. Since the listing-by-database is usually useless (the developers thinking that 127,824 results are preferable to eight), I leave it at ranked by purported relevance for the initial useless search. "Purported" said advisedly, since the initial search from trees in NewSearch to such a great extent disregards entered names, dates and places.

It's just that I noticed that this one's results were particularly wonky.

Re: Soundex is back bedeviling default NewSearch

Posted: 30 Jun 2012 3:13PM GMT
Classification: Query
> Unfortunately, all ~initial~ searches from tree
> individuals are initially global and either ranked or by
> database, depending on how one's prior search was
> configured. It is not possible to define the search
> parameters before doing the initial search-from-tree.

I fully realize that there are many, many successful approaches to using the ancestry.com search engine and that the successful approaches can vary widely. I've seen demos of ancestry.com in genealogy classes that look so different from what I do that I would swear that the instructor wasn't even using the same search engine that I use every day, and yet the instructor is getting good results from his or her searches.

Having said that, I'm not sure I understand the comments I quoted above about doing initial searches and setting up the initial search parameters. Let me describe what I do, and maybe you can explain things again by relating the approach I take to the approach you take.

First of all, I always use exact search. Well, maybe once a year I will try ranked search again after hearing a success story from someone who uses it, just to see what I'm missing or to see if maybe it's improved since the last time I tried. But so far, every time I try to kick the football again, Lucy jerks the football away at the last second as always. In other words, every time I try to use ranked search again the results are always an abject failure for me as always. Maybe some day that will change.

Second of all, I usually start with a global search. My basic opinion is that the most effective searches are against specific databases, not global searches. So why do I usually start with a global search? It's because I always use the option to display the results "Summarized by category". So the purpose of my initial search is not to find a list of individuals. The purpose of my initial search is to find a list of databases. And the databases can vary from things like census images and courthouse marriage record images all the way to trees compiled by ancestry.com members - all from the same initial search.

For example, the state I search the most is Tennessee and I have the great, good fortune that ancestry.com has a database of images of courthouse marriage records for most Tennessee counties. The easiest way to get there is to do a global search for John Doe (or whatever his or her name is) with some number of filters. Depending on the person's name, I might use wildcards or some of the "phonetic" or "similar" options for the name. I would specify that the event was a marriage in Tennessee, and often I would give an estimated year, +-5 or +-10. By the way, New Search really needs +-20, which is something that Old Search had. And for this search, that's about it. Such a search will usually yield a few hits in the Tennessee marriages database and a few hits in the trees.

I could give numerous other examples, and the exact setup of the search is essentially the same no matter what you are looking for. The same idea works to find data in trees, to find census information, to find death records, etc.

My technique doesn't work well to find data that has been OCR'd and indexed by keyword. But I don't think any technique works well for such data. Newspaper collections are darn near unsearchable with any technique. City Direcory collections are searchable, but it's extremely hard. You pretty much have to know the exact city and year, and go to that particular city directory.

I said I always do exact search and that I usually start with a global search (probably more than 95% of the time). The exception on starting with a global search is that sometimes it's better to go to the Card Catalog and find a particular database. But that doesn't change the philosophy. The philosophy is to find one or more databases and search within them with exact search. The initial global search is one way to find a list of databases within which to search. The Card Catalog is another way to find a list of databases within which to search.

Jerry

Re: Soundex is back bedeviling default NewSearch

Posted: 30 Jun 2012 5:51PM GMT
Classification: Query
Edited: 30 Jun 2012 5:55PM GMT
"First of all, I always use exact search . . . ."

Jerry, while your search advice is sound, my initial search was for a tree person, and it was the initial results I was referring to in the OP.

To search from a tree person one has absolutely no option as to what is searched or in what manner when you click "search records" under the individual person's main name/vitals box on the overview page (except as to whether your default is in NewSearch or OldSearch, which is not selectable from a tree individual display either).

One must do the initial junk-global search before getting access to the ~database group filters list~ and the ~search form~ that allows selecting search parameters.

Cheers,
Jade
per page

Find a board about a specific topic