Overview | ProBase | Snapshots | DF-ITF | Evaluation

Our analysis on the Bing search log during the period of September of 2007 to June of 2009 shows that about 62% of the queries contain at least one concept term. More detailed analysis revealed that common web queries can be classified into a number of different patterns. The following five basic patterns account for the majority of all the Bing queries during that period:

1. Single Entity (E)
2. Single Concept (C)
3. Single Entity + Attributes (E+A)
4. Single Concept + Attributes (C+A)
5. Single Concept + Keywords (C+K)

These patterns can be combined to form more complex patterns. In the paper, we focus on one of them:

Concept + Keywords + Concept (C+K+C)

To evaluate the performance of online query processing, we create a set of benchmark queries that contain concepts, entities, and attributes, for example, "politicians commit crimes" (C+K+C), "large companies in chicago" (C+K), "president washington quotes" (E+A), etc. The first 6 tables show the queries we used. 10 queries for each pattern. (E), (C), (E+A) and (C+A) queries are randomly selected from Bing's search log, so the rank column shows their rankings by frequency.

One assumption we made in the paper is that we can estimate the association between an entity term and a keyword using simple two-way word association. We described this in detail in the paper. Here we take the 10 benchmark C+K queries, substituting the concept for each of them to generate a set of E+K queries, and highlight the pivot word we found in some of these E+K queries. The last table shows them.

• Single Entity (E)

  These 10 queries are randomly selected from Bing's two-year search log. Freq. is the corresponding query's frequency and Ranking is its frequency ranking. (The same as below )

  #	E Queries	Freq.	Ranking
  E-1	(house beautiful)	27285	83899
  E-2	(borland)	15628	146134
  E-3	(witco)	2366	911523
  E-4	(hicksville)	1408	1490513
  E-5	(alan taylor)	751	2691832
  E-6	(condobolin)	654	3060796
  E-7	(george low)	216	8569491
  E-8	(pigmy love circus)	199	9235273
  E-9	(still bill)	117	14965256
  E-10	(kip hanrahan)	99	17425678

• Single Concept (C)

  #	C Queries	Freq.	Ranking
  C-1	[cars]	2286411	528
  C-2	[online services]	23729	96622
  C-3	[american artists]	12421	183182
  C-4	[boutiques]	10815	209807
  C-5	[classic fairy tales]	7154	314209
  C-6	[british authors]	3311	662012
  C-7	[red sox players]	1404	1493864
  C-8	[italian composers]	716	2813231
  C-9	[media conglomerates]	297	6384077
  C-10	[mainstream movies]	105	16564335

• Single Entity + Attributes (E+A)

  #	E+A Queries	Freq.	Ranking
  EA-1	(pedro infante) <music>	658	3043603
  EA-2	(franklin) <time>	84	20166776
  EA-3	(assurant) <employees>	70	23591037
  EA-4	(king arthur) <music>	58	27870316
  EA-5	(phil harris) <age>	30	50799136
  EA-6	(david beckham) <quote>	28	53243346
  EA-7	(chennai) <place>	17	85110446
  EA-8	(president washington) <quotes>	16	87769744
  EA-9	(aaron neville) <album>	10	132744902
  EA-10	(peanuts) <artist>	10	132744902

• Single Concept + Attributes (C+A)

  #	C+A Queries	Freq.	Ranking
  CA-1	[famous people] <birthdays>	2116	1012793
  CA-2	[common allergies] <symptoms>	1424	1474277
  CA-3	[movies] <quotes>	853	2389110
  CA-4	[professional boxers] <champions>	145	12310179
  CA-5	[violinists] <tool>	28	52969723
  CA-6	[films] <soundtracks>	21	68772582
  CA-7	[images] <cameras>	21	68772582
  CA-8	[herbal supplements] <energy>	13	109177365
  CA-9	[nba players] <position>	12	113769727
  CA-10	[funds] <home page>	11	126590746

• Single Concept + Keywords (C+K)

  #	C+K Queries
  CK-1	[east asian countries] with nuclear capability
  CK-2	[american cities] sigmod
  CK-3	[large companies] in chicago
  CK-4	las vegas [outdoor activities]
  CK-5	[international organizations] focus on environmental protection
  CK-6	horse [medical conditions]
  CK-7	[name brands] in chinese market
  CK-8	[football players] own goal
  CK-9	[astronauts] fly to the moon
  CK-10	[famous people] bribery

• Concept + Keywords + Concept (C+K+C)

  #	C+K+C Queries
  CKC-1	[companies] buy [tech companies]
  CKC-2	[politicians] commit [crimes]
  CKC-3	[extreme sports] in [asian countries]
  CKC-4	[database conferences] in [european cities]
  CKC-5	[presidents] graduated from [universities]
  CKC-6	[rivers] flow into [seas]
  CKC-7	[actors] marry [actresses]
  CKC-8	[football stars] join [football teams]
  CKC-9	[cars] owned by [celebrities]
  CKC-10	[peoples] believe in [religions]

• Pivot words

  #	Pivot Word in Each Query
  CK-1	(pr china) with nuclear capability
  	(republic korea) with nuclear capability
  	(north korea) with nuclear capability
  CK-2	(san diego) sigmod
  	(washington dc) sigmod
  	(ann arbor)1 sigmod
  CK-3	(general electric) in chicago
  	(american express) in chicago
  	(microsoft corp) in chicago
  CK-4	las vegas (american football)
  	las vegas (water sport)
  	las vegas (nascar race)
  CK-5	(family health international) focus on ...
  	(world health organization) focus on ...
  	(the pan american health organization) focus on ...
  CK-6	horse (high blood pressure)
  	horse (low blood sugar)
  	horse (skin cancer)
  CK-7	(new balance) in chinese market
  	(calvin klein) in chinese market
  	(hewlett packard) in chinese market
  CK-8	(paul robinson) own goal
  	(graham alexander) own goal
  	(jonathan woodgate) own goal
  CK-9	(john glenn) fly to the moon
  	(neil armstrong) fly to the moon
  	(michael collins) fly to the moon
  CK-10	(james brown) bribery
  	(george bush) bribery
  	(bill clinton) bribery

  ¹Both are pivot words.