Nestoria - tech demo

Marc Tobias Metten

Lokku Ltd.

start-up

• started Apr/2006
• 9 total, 4.742 engineers
• Covent Garden (London)

What we do

property for sale in Oxford

• vertical search for houses/flats
  • in = data feeds, crawling
  • out = quality ranked results
  • no logins
  • no banner ads

Homepage

Result page

technical challenges

• geographic (spatial) data
• geocoding
• query/document analysis
• ranking
• search engine marketing

geographic data

norton

• 50.000 locations UK
• 80.000 locations Spain
• plus
  • local labels, e.g. Soho
  • DLR stations, islands, golf courses...
  • Basque, Galician ... 'Almazora' vs 'Almassora'
• disambiguation
  • 23x borough 'Norton'
  • 5x 'Waterloo'
  • 'Bilbao' (city) vs 'Bilbao' (barrio de Madrid)

geocoding

• public (Google,Yahoo) geocoders too slow, limit transactions
• we get lots of 'bad' data
• talk: Geocoding in the UK

• Zaragoza / Utebo
• Conde De Vistahermosa, 34, 3 Izq.
• Fermin Izquierdo, Pilar, Fuencarral - El Pardo, Madrid Capital Cp 28029 (Madrid)
• Castellon/Castello, Almazora/Almassora, AVDA. CONSTITUCION
• Marques De Teverga, Cp 33005, Oviedo (Centro) Cp 33005 (Asturias)
• Caceres / Ocho De Marzo
• 9 Calle de las Maravillas

query/document analysis

• spell-checking
  Isle of White => Isle of Wight
Notthing Hall => Notting Hill
Thirlstane => Thurlstone
  
• document classification
• text and keyword extraction
  Stantons are pleased to offer for sale this end terrace property located in the ever popular residential area of Shire Lodge. The property has undergone many improvements over recent years and is considered to be in good order throughout. The accommodation comprises of three first floor bedrooms, first floor bathroom, entrance hall, lounge/dining room, a fitted kitchen with separate utility room, UPVC conservatory and front and rear gardens. The property further benefits from a single garage, UPVC double glazing and gas fired central heating to radiators. The property is considered to be an ideal first time purchase.

  ==> conservatory, garage, garden, heating, kitchen, lodge, terrace

duplicates

• up to 50% duplicates on 1 mio results
  • multiple agents list on multiple portals
  • multiple apartments per house/block
  • new constructions

• incomplete data
• "nearby" - fussy
• "about same price" - fussy
• "same photo" - too CPU intensive

ranking

• result quality
• geographic proximity
  • also confidence/accuracy
• relevance to session context
• result recommendation
  • "houses similar to this"
  • based on past user behavior

search engine marketing

• several hundred thousand keywords
  • dozen accounts
  • exclusion flags
• geographic focus
• geographic campaigns
• "auto-optimizing"

API

    [...]
    {
        "property_type" : "flat",
        "listing_type" : "buy",
        "num_beds" : "1",
        "metadata_name" : "avg_1bed_flat_buy_monthly",
        "data" : {
           "2007_m10" : {
              "avg_price" : "524995.78",
              "datapoints" : "15"
           }
        },
    },
    [...]
    

• data
  • listings, lots of filters
  • avg house prices
• format
  • JSON, Javascript with callback
  • XML
  • geoRSS, KML
• wrappers
  • Perl (on CPAN)
  • PHP (on PEAR)
  • Python (an Egg)

API uses

houseprice widget


OpenStreeMap on Garmin

• Facebook App
• Yahoo Widget
• Javascript widgets
  • list of properties
  • small Google Map
  • avg. house prices
• Ideas...
  • more historical data, charts
  • heatmaps
  • alternative interfaces: iphone version...
  • desktop client
  • Flash games

API example

API example

API example

Density of homes for sale across the UK

Questions?