WO2012109959A1 - 检索词的聚类方法和装置 - Google Patents

检索词的聚类方法和装置 Download PDF

Info

Publication number
WO2012109959A1
WO2012109959A1 PCT/CN2012/070824 CN2012070824W WO2012109959A1 WO 2012109959 A1 WO2012109959 A1 WO 2012109959A1 CN 2012070824 W CN2012070824 W CN 2012070824W WO 2012109959 A1 WO2012109959 A1 WO 2012109959A1
Authority
WO
WIPO (PCT)
Prior art keywords
search term
search
clustering
candidate
term
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Ceased
Application number
PCT/CN2012/070824
Other languages
English (en)
French (fr)
Inventor
赫南
王迪
郭阳
胡立新
王艳敏
朱建朋
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Tencent Technology Shenzhen Co Ltd
Original Assignee
Tencent Technology Shenzhen Co Ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Tencent Technology Shenzhen Co Ltd filed Critical Tencent Technology Shenzhen Co Ltd
Priority to US14/000,083 priority Critical patent/US20140019452A1/en
Publication of WO2012109959A1 publication Critical patent/WO2012109959A1/zh
Anticipated expiration legal-status Critical
Ceased legal-status Critical Current

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
    • G06F16/35—Clustering; Classification
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q30/00—Commerce
    • G06Q30/02—Marketing; Price estimation or determination; Fundraising
    • G06Q30/0241—Advertisements
    • G06Q30/0251—Targeted advertisements
    • G06Q30/0255—Targeted advertisements based on user history
    • G06Q30/0256—User search
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
    • G06F16/33—Querying
    • G06F16/3331—Query processing
    • G06F16/3332—Query translation
    • G06F16/3338—Query expansion

Definitions

  • the invention relates to a network search technology, in particular to a clustering method and device for searching words. Background of the invention
  • search term may be specifically identified as an advertisement advertisement provided by the advertiser, and may also be referred to as a purchase word, so as to facilitate the user to search for the corresponding advertisement through the search term.
  • the process of clustering search terms can be abstracted into a process of clustering a set of short text strings.
  • the most commonly used clustering method is: For a search term provided by an advertiser, only the term, the search term provided by the advertiser and the found search term are clustered together. Thus, when the search engine user retrieves the corresponding advertisement through a search term, the advertisement corresponding to the search term and the advertisement corresponding to the search term clustered by the search term are displayed to the user.
  • search terms although the advertiser does not provide it, but it is substantially related to the advertisement corresponding to the search term provided by the advertiser, and the aforementioned clustering method is to perform only the word-related clustering of the search terms provided by the advertiser. , which does not take into account these other search terms related to the search term provided by the advertiser and which have not yet been provided by the advertiser, which reduces the accuracy of clustering of the search terms.
  • the present invention provides a clustering method and apparatus for search terms to improve the accuracy and relevance of clustering of search terms.
  • a clustering method for search terms including:
  • a clustering operation is performed on the first search term in the candidate search term set and the second search term associated with the first search term based on text features and/or semantic features of the search term.
  • a clustering device for searching words comprising:
  • a establishing unit configured to establish a candidate search term set, where the candidate search term set includes a first search term provided by a user, and a second search term related to the first search term;
  • a clustering unit configured to perform a clustering operation on the first search term in the set of candidate search words and the second search term related to the first search term according to text features and/or semantic features of the search term.
  • the clustering method and apparatus for searching for a search term provided by the present invention does not cluster the search terms provided by the user only by the user, as in the prior art. Considering simultaneously the search term provided by the user, and other search terms related to the search term provided by the user, and the search term provided to the user according to the text feature and/or semantic feature of the search term, and the search term provided by the user The related other search terms are clustered, thereby increasing the accuracy and relevance of the search term clustering.
  • FIG. 1 is a basic flowchart of an embodiment of the present invention
  • FIG. 2a is a flowchart of step 102 according to an embodiment of the present invention
  • FIG. 2b is a flowchart of potential clustering relationship mining according to an embodiment of the present invention
  • FIG. 3 is a first schematic diagram of a topology structure between search terms according to an embodiment of the present invention
  • FIG. 3b is a second schematic diagram of a topology structure between search terms according to an embodiment of the present invention
  • FIG. 3 is a third schematic diagram of a topology structure when a search term is added according to an embodiment of the present invention
  • FIG. 4 is a schematic diagram of a new search term provided by an embodiment of the present invention
  • FIG. 5 is a basic structural diagram of an apparatus according to an embodiment of the present invention.
  • FIG. 6 is a detailed structural diagram of an apparatus according to an embodiment of the present invention. Mode for carrying out the invention
  • the present invention does not cluster only the search terms provided by the user, such as an advertiser, as in the prior art, but provides the user with the text features and/or semantic features of the search terms.
  • the search term, and the search term cluster associated with the search term are added to increase the accuracy of clustering of the search terms.
  • the method provided by the present invention is described below.
  • FIG. 1 is a basic flowchart of an embodiment of the present invention. As shown in Figure 1, the process can include the following steps:
  • Step 101 Establish a candidate search term set, where the candidate search term set includes a first search term provided by a user, and a second search term related to the first search term.
  • the second search term related to the first search term provided by the user may specifically include the following two methods or any one of the following manners: mode 1, determining a search term that matches the first search term provided by the user, Determining the determined search term as a second search term related to the first search term provided by the user; mode 2, searching for the first search term provided by the user as a keyword, and determining the search term in the search result as The second term related to the first search term provided by the user Search terms.
  • the search term obtained by the method 1 may be: a search term obtained by performing a string conversion process on the first search term provided by the user; or a first search term determined according to actual experience.
  • the search term used together for example, if the first search term provided by the user is a coffee maker, it is known from experience that the coffee maker is usually used frequently with a coffee cup or the like, based on which it can be determined that the coffee maker provided by the user matches.
  • the search term can be a coffee cup or the like.
  • the search term obtained by the method 2 may specifically be: searching for the first search term provided by the user as a keyword, and the search term in the obtained search result.
  • the search may be specifically implemented by a user search string and a search term mapping integration system (QBM: Query Bidterm Merge), wherein the QBM implementation may be: searching with the first search term provided by the user as input, from the search.
  • QBM search term mapping integration system
  • the QBM implementation may be: searching with the first search term provided by the user as input, from the search
  • the search term is obtained from the search result, and the obtained search term is used as a search term related to the first search term provided by the user.
  • the candidate search term set can be obtained through step 101. It should be noted that, in this embodiment, it is necessary to ensure that there are no duplicate search terms in the candidate search term set obtained in step 101.
  • Step 102 Perform a clustering operation on the first search term in the candidate search term set and the second search term related to the first search term according to the text feature and/or the semantic feature of the search term.
  • the first search term and the second search term related to the first search term in the candidate search term set may be calculated according to the text feature and/or the semantic feature of the first search term.
  • the similarity value, the first search term and the second search term having a higher similarity value with the first search term are clustered together.
  • the step 102 can be embodied by the flow shown in Figure 2a.
  • FIG. 2a is a flowchart of step 102 according to an embodiment of the present invention.
  • the flow shows the specific implementation principle of the basic clustering relationship, as shown in FIG. 2a, the process may include Next steps:
  • Step 201a Calculate a similarity value between the first search term and each of the related second search terms according to the text feature and/or the semantic feature of the first search term.
  • Step 202a If the similarity value between the first search term and the second search term is greater than or equal to the first preset threshold, the first search term and the second search term are clustered together.
  • step 202a the first search term and the second search term associated with the first search term having a similarity value greater than or equal to the first preset threshold may be clustered together, that is, the present search word is implemented.
  • the present search word is implemented.
  • the embodiment further provides a mining process of the potential clustering relationship, which can be embodied by the process shown in FIG. 2b.
  • FIG. 2b is a flowchart of potential clustering relationship mining according to an embodiment of the present invention. As shown in Figure 2b, the process can include the following steps:
  • Step 201b Select, from each of the second search terms related to the first search term, a second search term whose similarity value with the first search term is greater than or equal to a second preset threshold.
  • the step 201b may be further replaced by: selecting and selecting the second search term from the first search term.
  • a second search term whose similarity value between the search terms is greater than or equal to the second predetermined threshold.
  • the second preset threshold in the step 201b is independent of the first preset threshold in step 202a, and the two may be equal or unequal.
  • Step 202b calculating a similarity value between the selected two second search terms, and if the calculated similarity value is greater than or equal to the first preset threshold, clustering the two second search terms Together.
  • the first search term and the second clustered together in step 202a are used in the embodiment of the present invention.
  • a search term ie, a cluster relationship between the first search term and the second search term
  • the second search terms clustered together in 202b are combined to form a full-scale clustering result of the embodiment of the present invention.
  • the clustering of step 202a and the clustering of step 202b may be implemented according to a similar machine learning model, which is not specifically limited herein.
  • the first search terms provided by the user are bl, b3, b4 and b5 respectively, wherein, by step 101, it can be obtained that: the second search terms related to bl are b2, b3 and b4, and the second search term related to b3 For b5, b6 and b4, the second search terms associated with b4 are b7, b8 and b9, and the second search term associated with b5 is b3. All search terms are represented by the graph data structure shown in Figure 3a. Referring to FIG. 3a, FIG. 3a is a first schematic diagram of a topology structure between search terms according to an embodiment of the present invention. In Fig.
  • each search term is taken as an arrow of node bi (i takes a value of 1 to 9), from node bi to node bj (j takes a value of 1 to 9), indicating that bi can be expanded to bj, that is, , the related term with bi is bj.
  • the topology shown in Fig. 3a is a directed acyclic graph, that is to say, the correlation between the two search terms is not guaranteed to be bidirectional, specifically:
  • the search term related to bi is the search term bj, but the search term bj does not necessarily extend the search term related to the search term bj as the search term bio
  • step 201a it can be obtained that: for bl, the similarity value wl2 between bl and b2, the similarity values wl3, bl and b4 between bl and b3 are calculated according to the text feature and/or semantic feature of bl.
  • the similarity value wl4; for b3, the similarity value wl4 between b3 and b4, the similarity value between b3 and b5, the similarity between b3 and b6 is calculated according to the text feature and/or semantic feature of b3 Degree value w36; for b4, calculate the similarity value w47 between b4 and b7 according to the text feature and/or semantic feature of b4, the similarity value w48 between b4 and b8, the similarity value w49 between b4 and b9 ; for b5, according to the text characteristics of b5 and / or semantic feature calculates the similarity value w53 between b5 and b3.
  • FIG. 3b is a second schematic diagram of a topological structure between search terms provided by an embodiment of the present invention.
  • Figure 3b shows the clustering relationship between the interconnected search terms, wherein the two search terms connected by the solid line indicate that the two search terms have a cluster relationship: the two are considered equivalent and can be clustered Together; the two search terms connected by the dotted line have the clustering relationship: The two are not equivalent, and cannot be clustered together, and the dotted line can be removed later.
  • the second search terms with bl are: b2, b3, and b4, so, based on step 201b, when the similarity values between b2, b3, and b4 and bl are greater than or equal to
  • the present invention can supplement three potential clustering relationships: clustering relationship between b2 and b3, clustering relationship between b2 and b4, and clustering between b3 and b4 relationship.
  • the clustering relationship between b3 and b4 has been determined in the above step 202a.
  • the present invention may omit the operation of determining the clustering relationship between b3 and b4, It is necessary to increase the clustering relationship between b2 and b3 and the clustering relationship between b2 and b4. Then calculate the similarity value between b2 and b3, and the similarity value between b2 and b4, determine whether the cluster relationship between b2 and b3 and the cluster relationship between b2 and b4 meet the criteria of clustering.
  • the clustering relationship between the search terms is represented by a solid line (also referred to as an edge relationship) between the search terms. Therefore, the embodiment of the present invention can only traverse the edge relationship, so that the present invention can be made.
  • the complexity of the embodiment is reduced to 0 (n + e), where n represents the number of search terms and e represents the number of edge relationships.
  • the second search term related to the first search term provided by the user in FIG. 3a may be further mined, and the second search term is N (for example, N is 3)
  • N for example, N is 3
  • the potential clustering relationship between the "children" nodes within the hop For the specific implementation, refer to the process shown in Figure 2b, which will not be described in detail here.
  • the set of candidate search terms is not fixed, and the search terms can be incremented over time.
  • the candidate search term set newly adds the first search term provided by the user, and the newly added first search term is newly appearing relative to all previous search terms.
  • the newly added first search term it is also necessary to perform a clustering operation similar to that shown in Fig. 2a and Fig. 2b, and at the same time, integrate the result obtained after performing the clustering operation with the previous clustering result. See the process shown in Figure 4.
  • FIG. 4 is a flowchart of a process of adding a first search term (indicated as an incremental update process) according to an embodiment of the present invention.
  • the process may include the following steps: Step 401: Determine a second search term related to the added first search term, and compare the added first search term with the determined second search term related to the added first search term. A second search term different from any one of the candidate search term sets is added to the candidate search term set.
  • the search terms stored in the candidate search term set before the execution of step 401 are bl to b9 shown in Fig. 3a, and when executed to this step 401, if the following two first search words are newly added: nl and n2.
  • the second search terms related to nl are b5 and b6, and the second search terms related to n2 are bl, b2, b3, b4, b8, and n3, as shown in FIG. 3d. Since b5 and b6 associated with nl, and bl, b2, b3, b4, b8 associated with n2 are already stored in the set of candidate search terms, this step 401 can only refer to nl, n2, and n2. N3 is added to the set of candidate terms.
  • Step 402 Perform a clustering operation on the newly added first search term in the candidate search term set and the second search term related to the first search term according to the text feature and/or the semantic feature of the search term.
  • step 402 is described by taking the newly added first search term as nl as an example, and the added other search terms are similar in principle.
  • step 402 when performing this step 402, based on the flow shown in FIG. 2a, the similarity value between n1 and b5 is calculated according to the text feature and/or the semantic feature of n1, and the similarity value between n1 and b6 is calculated. Then, it is determined whether the similarity value between nl and b5 is greater than or equal to the first preset threshold, and if so, it is determined that nl and b5 are equivalent, and the two can be clustered together, otherwise, nl and b5 are not clustered. Together. The same operation is performed for the similarity value between nl and b6.
  • Step 403 Perform mining of a potential clustering relationship on the second search term related to the added first search term in the candidate search term set.
  • the process of the potential clustering relationship may be performed by using the process shown in FIG. 2b, and the bill is described as: each second search term related to the added first search term from the candidate search term set, or increased from And selecting, from each of the second search words, the first search term, the second search term having a similarity value with the first search term greater than or equal to a second preset threshold; calculating any two selected And a similarity value between the two search terms, if the calculated similarity value is greater than or equal to the first preset threshold, the two second search terms are clustered together.
  • the search term nl As an example, since it is determined in step 401 that the second search term related to the n1 is b5 and b6, when performing this step 403, if b5 and b6 respectively If the similarity value between n1 and n1 is greater than the second preset threshold, the similarity value between b5 and b6 may be calculated, and if the calculated similarity value is greater than or equal to the first preset threshold, the two are The search terms b5 and b6 are clustered together, otherwise b5 and b6 are not clustered together.
  • Incremental clustering results the cluster relationship between the newly added first search term (indicated as an incremental search term) and the original existing search term (recorded as the old search term) is realized (hereinafter referred to as Incremental clustering results).
  • the incremental clustering result and the pre-existing full-quantity clustering result are collectively referred to as the final clustering result of the present invention.
  • the second search term related to the first search term is not fixed, and the search term is also changed according to the user, and the method provided by the embodiment of the present invention should also be able to Reflect this change.
  • the change is implemented by periodically updating the set of candidate search terms (referred to as full update), and is specifically implemented as: when the set full amount update time arrives, determining, for the first search term in the candidate search term set, the first Searching for a second search term related to the word, placing the first search term and the determined second search term related to the first search term into a new candidate search term set, and then according to FIG. 2a and FIG. 2
  • the illustrated process clusters the search terms in the new candidate search term set to obtain a full-scale clustering result. This Can be described by the image of Table 1.
  • the corresponding QBM extension result of the first search term is Q ⁇ i B, and the extended result is mainly a set of second search terms related to the first search term.
  • the clustering result obtained by clustering the first search term and the second search term based on the flow shown in Fig. 2a and Fig. 2b is: CF CXB);
  • the full update starts on the ith day, the end of the kth day, and on the k+1th (ie, L) day, the synchronous operation of the full amount of data and the incremental data is performed, that is, the kth All of the first search terms in the +1 (i.e., L) day candidate search term set perform the flow shown in FIG.
  • the device provided by the embodiment of the present invention is described below.
  • FIG. 5 is a basic structural diagram of an apparatus according to an embodiment of the present invention. As shown in Figure 5, the device can include:
  • the establishing unit 501 is configured to establish a candidate search term set, where the candidate search term set includes a first search term provided by the user, and a second search term related to the first search term; and a clustering unit 502, configured to perform the search according to the search
  • the textual feature and/or semantic feature of the word performs a clustering operation on the first search term in the set of candidate search terms and the second search term associated with the first search term.
  • the device shown in FIG. 5 can be specifically seen in FIG. 6.
  • FIG. 6 is a detailed structural diagram of an apparatus according to an embodiment of the present invention.
  • the apparatus may include an establishing unit 601 and a clustering unit 602, wherein the establishing unit 601 and the clustering unit 602 have functions similar to the establishing unit 501 and the clustering unit 502 shown in FIG. 5, respectively. No longer.
  • the apparatus may further include:
  • An adding unit 603, configured to: when the user adds a new first search term, determine a second search term related to the added first search term, and add the added first search term and the determined a second search term different from any one of the candidate search term sets in the second search term related to the first search term is added to the candidate search term set;
  • the clustering unit 602 is further configured to perform, according to the text feature and/or the semantic feature of the search term, the newly added first search term in the candidate search term set and the second search term related to the first search term. Clustering operation.
  • the apparatus further includes:
  • the updating unit 604 is configured to: when the set full amount update time arrives, determine, for the first search term in the candidate search term set, a second search term related to the first search term, the first search term And determining the second search term associated with the first search term into one A new set of candidate search terms.
  • the clustering unit 602 is further configured to perform clustering on the first search term in the new candidate search term set and the second search term related to the first search term according to the text feature and/or the semantic feature of the search term. operating.
  • the clustering unit 602 performs a clustering operation through the following subunits:
  • a calculating subunit 6021 configured to calculate, according to a text feature and/or a semantic feature of the first search term, a similarity value between the first search term and each second search term related to the first search term;
  • the clustering sub-unit 6022 is configured to cluster the first search term and the second search term when the similarity value between the first search term and the second search term is greater than or equal to the first preset threshold .
  • the clustering sub-unit 6022 is further configured to select the first search term from each of the second search terms related to the first search term, or from each of the second search terms clustered with the first search term. a second search term having a similarity value greater than or equal to a second predetermined threshold; and calculating a similarity value between the selected two second search terms, if the calculated similarity value is greater than or equal to The first preset threshold is clustered, and the two second search terms are clustered together, and the first preset threshold is independent of the second preset threshold.
  • the clustering method and apparatus for searching for a search term provided by the present invention does not cluster the search terms provided by the user only by the user, as in the prior art. Considering simultaneously the search term provided by the user, and other search terms related to the search term provided by the user, and the search term provided to the user according to the text feature and/or semantic feature of the search term, and the search term provided by the user Clustering related related terms, which obviously increases the accuracy of clustering of search terms;
  • the present invention also excavates various first related to the first search term provided by the user.
  • the clustering relationship between the two search terms compared with the prior art, the clustering relationship between the search terms can be deeply explored, and the clustering of the search words is more accurate.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Business, Economics & Management (AREA)
  • General Physics & Mathematics (AREA)
  • Physics & Mathematics (AREA)
  • Accounting & Taxation (AREA)
  • Development Economics (AREA)
  • Finance (AREA)
  • Strategic Management (AREA)
  • General Engineering & Computer Science (AREA)
  • Databases & Information Systems (AREA)
  • Data Mining & Analysis (AREA)
  • Entrepreneurship & Innovation (AREA)
  • General Business, Economics & Management (AREA)
  • Marketing (AREA)
  • Economics (AREA)
  • Game Theory and Decision Science (AREA)
  • Computational Linguistics (AREA)
  • Information Retrieval, Db Structures And Fs Structures Therefor (AREA)

Description

检索词的聚类方法和装置 本申请要求于 2011年 02月 18日提交中国专利局、 申请号为 201110043030.7、 发明名称为 "检索词的聚类方法和装置" 的中国专利 申请的优先权， 其全部内容通过引用结合在本申请中。 技术领域
本发明涉及网络搜索技术， 特别涉及检索词的聚类方法和装置。 发明背景
在网络搜索技术中， 用户都是通过检索词搜索到相应的结果。 应用 于竟价广告系统中， 该检索词具体实现时可为广告商提供的广告的标 识，也可称为购买词， 目的是便于用户通过该检索词搜索到相应的广告。
在竟价广告系统中， 为了提高广告展现质量， 需要对广告商提供的 检索词进行聚类。 其中， 对检索词进行聚类的过程可以抽象为对一个短 文本串的集合进行聚类的过程。
目前， 最常用的聚类方法为： 针对一广告商提供的检索词， 仅从目 词， 将该广告商提供的检索词和找出的检索词聚类在一起。 如此， 当搜 索引擎用户通过一检索词检索相应的广告时， 将与该检索词对应的广 告， 以及与该检索词聚类在一起的检索词对应的广告展示给用户。
然而， 有一些检索词， 尽管广告商没有提供， 但其实质上与广告商 提供的检索词对应的广告相关， 而前述的聚类方法是仅将广告商提供的 检索词进行字面相关的聚类， 没有考虑到这些与广告商提供的检索词语 义相关的、 且目前还未被广告商提供的其他检索词， 这降低了检索词聚 类的准确度。 发明内容
本发明提供了检索词的聚类方法和装置， 以提高检索词聚类的准确 度和相关度。
本发明提供的技术方案包括：
一种检索词的聚类方法， 包括：
建立候选检索词集合， 所述候选检索词集合包含由用户提供的第一 检索词、 以及与第一检索词相关的第二检索词；
根据检索词的文本特征和 /或语义特征对所述候选检索词集合中的 第一检索词和与该第一检索词相关的第二检索词执行聚类操作。
一种检索词的聚类装置， 包括：
建立单元， 用于建立候选检索词集合， 所述候选检索词集合包含由 用户提供的第一检索词、 以及与第一检索词相关的第二检索词；
聚类单元，用于根据检索词的文本特征和 /或语义特征对所述候选检 索词集合中的第一检索词和与该第一检索词相关的第二检索词执行聚 类操作。
由以上技术方案可以看出， 本发明提供的检索词的聚类方法和装置 在进行检索词聚类时， 并非像现有技术那样仅将用户提供的检索词进行 字面关系的聚类， 而是同时考虑到用户提供的检索词， 以及与该用户提 供的检索词相关的其他检索词，并根据检索词的文本特征和 /或语义特征 对用户提供的检索词， 以及与该用户提供的检索词相关的其他检索词进 行聚类， 因而能够增加检索词聚类的准确度和相关度。 附图简要说明
图 1为本发明实施例提供的基本流程图；
图 2a为本发明实施例提供的步骤 102的流程图； 图 2b为本发明实施例提供的潜在聚类关系挖掘流程图；
图 3a为本发明实施例提供的检索词之间的拓朴图结构第一示意图； 图 3b为本发明实施例提供的检索词之间的拓朴图结构第二示意图； 图 3c为本发明实施例提供的检索词之间潜在的聚类关系示意图； 图 3d为本发明实施例提供的增加检索词时拓朴图结构第三示意图； 图 4为本发明实施例提供的新增加检索词时的流程图；
图 5为本发明实施例提供的装置的基本结构图；
图 6为本发明实施例提供的装置的详细结构图。 实施本发明的方式
为了使本发明的目的、 技术方案和优点更加清楚， 下面结合附图和 具体实施例对本发明进行详细描述。
本发明在进行检索词聚类时， 并非像现有技术那样仅将用户比如广 告商提供的检索词进行字面关系的聚类， 而是根据检索词的文本特征和 /或语义特征将用户提供的检索词， 以及与该检索词相关的检索词聚类， 以便增加检索词聚类的准确度， 下面对本发明提供的方法进行描述。
参见图 1 , 图 1为本发明实施例提供的基本流程图。 如图 1所示， 该流程可包括以下步骤：
步骤 101 , 建立候选检索词集合， 所述候选检索词集合包含由用户 提供的第一检索词、 以及与第一检索词相关的第二检索词。
本步骤 101中， 与用户提供的第一检索词相关的第二检索词具体可 包括以下两种方式或任一方式确定： 方式 1 , 确定与该用户提供的第一 检索词匹配的检索词， 将该确定的检索词确定为与用户提供的第一检索 词相关的第二检索词； 方式 2, 以该用户提供的第一检索词为关键词搜 索， 将搜索结果中的检索词确定为与用户提供的第一检索词相关的第二 检索词。
其中， 方式 1得到的检索词， 具体可为： 通过对该用户提供的第一 检索词进行筒单的字符串变换处理所得到的检索词； 或者根据实际经验 确定出的与第一检索词经常在一起使用的检索词， 比如， 如果用户提供 的第一检索词为咖啡壶， 则根据经验可以知道咖啡壶通常与咖啡杯等经 常使用， 基于此， 可确定与该用户提供的咖啡壶匹配的检索词可为咖啡 杯等。
其中， 方式 2得到的检索词， 具体可为： 以用户提供的第一检索词 为关键词进行搜索， 所得到的搜索结果中的检索词。 其中， 该搜索具体 可通过用户搜索串与广告检索词映射整合系统 （ QBM: Query Bidterm Merge ) 实现， 其中， QBM具体实现时可为： 以用户提供的第一检索词 作为输入进行搜索， 从搜索到的搜索结果中获取检索词， 将该获取的检 索词作为与该用户提供的第一检索词相关的检索词。
至此， 通过步骤 101即可得到候选检索词集合。 需要说明的是， 本 实施例需要保证步骤 101得到的候选检索词集合中没有重复的检索词。
步骤 102, 根据检索词的文本特征和 /或语义特征对所述候选检索词 集合中的第一检索词和与该第一检索词相关的第二检索词执行聚类操 作。
本步骤 102具体实现时，可根据第一检索词的文本特征和 /或语义特 征计算该第一检索词和所述候选检索词集合中与该第一检索词相关的 第二检索词之间的相似度值， 将该第一检索词和与该第一检索词具有较 高相似度值的第二检索词聚类在一起。 具体地， 该步骤 102可通过图 2a 所示的流程体现。
参见图 2a, 图 2a为本发明实施例提供的步骤 102的流程图。 该流 程示出了基本聚类关系的具体实现原理，如图 2a所示，该流程可包括以 下步骤：
步骤 201a, 根据第一检索词的文本特征和 /或语义特征计算该第一 检索词和其相关的每一个第二检索词之间的相似度值。
步骤 202a,如果该第一检索词和第二检索词之间的相似度值大于或 等于第一预设阈值， 则将该第一检索词和该第二检索词聚类在一起。
通过步骤 202a, 能够将第一检索词和其相关的、 且与该第一检索词 之间的相似度值大于或等于第一预设阈值的第二检索词聚类在一起， 即 实现了本发明实施例的基本聚类。
优选地， 为保证更加完整的聚类关系， 本实施例还提供了潜在聚类 关系的挖掘流程， 具体可通过图 2b所示的流程体现。
参见图 2b, 图 2b为本发明实施例提供的潜在聚类关系挖掘流程图。 如图 2b所示， 该流程可包括以下步骤：
步骤 201b,从与第一检索词相关的各个第二检索词中选取与该第一 检索词之间的相似度值大于或等于第二预设阈值的第二检索词。
作为本发明实施例的一种扩展， 为降低潜在聚类关系挖掘的复杂 度， 本步骤 201b还可替换为： 从与第一检索词聚类在一起的各个第二 检索词中选取与该第一检索词之间的相似度值大于或等于第二预设阈 值的第二检索词。
其中， 本步骤 201b中的第二预设阈值与步骤 202a中的第一预设阈 值无关， 两者可相等， 也可不等。
步骤 202b, 计算该选取的任意两个第二检索词之间的相似度值， 如 果该计算的相似度值大于或等于所述第一预设阈值， 则将该两个第二检 索词聚类在一起。
通过步骤 201b至步骤 202b, 能够实现潜在聚类关系的挖掘。
如此，本发明实施例将步骤 202a中聚类在一起的第一检索词和第二 检索词 （即该第一检索词和第二检索词之间具有聚类关系）， 以及步骤
202b中聚类在一起的第二检索词合并在一起，即可形成了本发明实施例 的全量聚类结果。 优选地， 本实施例中， 步骤 202a的聚类和步骤 202b 的聚类均可按照类似现有的机器学习模型实现， 这里并不具体限定。
为使图 2所示的流程更加清楚， 下面通过一个具体实施例对本发明 提供的流程进行描述。
假如用户提供的第一检索词分别为 bl , b3、 b4和 b5 , 其中， 通过 步骤 101 , 可以得到： 与 bl相关的第二检索词为 b2, b3和 b4, 与 b3 相关的第二检索词为 b5、 b6和 b4, 与 b4相关配的第二检索词为 b7、 b8和 b9, 与 b5相关的第二检索词为 b3。 将所有检索词通过图 3a所示 的图数据结构表示。 参见图 3a, 图 3a为本发明实施例提供的检索词之 间的拓朴图结构第一示意图。 在图 3a中， 将每个检索词作为节点 bi ( i 取值为 1至 9 ), 从节点 bi至节点 bj ( j取值为 1至 9 ) 的箭头， 表示 bi 可扩展出 bj , 也即， 与 bi的相关检索词为 bj。 从图 3a可以看出， 图 3a 示出的拓朴图是一个有向无环图， 也就是说， 两个检索词之间的相关关 系并非保证是双向相关， 具体为： 从 bi可以扩展出与 bi相关的检索词 为检索词 bj , 但从检索词 bj并非一定扩展出与检索词 bj相关的检索词 为检索词 bio
之后， 基于步骤 201a, 则可得到： 针对 bl , 根据 bl的文本特征和 / 或语义特征计算 bl和 b2之间的相似度值 wl2, bl和 b3之间的相似度 值 wl3 , bl和 b4之间的相似度值 wl4; 针对 b3 , 根据 b3的文本特征和 /或语义特征计算 b3和 b4之间的相似度值 wl4, b3和 b5之间的相似度 值 w35 , b3和 b6之间的相似度值 w36; 针对 b4, 根据 b4的文本特征和 /或语义特征计算 b4和 b7之间的相似度值 w47, b4和 b8之间的相似度 值 w48, b4和 b9之间的相似度值 w49; 针对 b5 , 根据 b5的文本特征和 /或语义特征计算 b5和 b3之间的相似度值 w53。
之后， 针对图 3a中用户提供的每一个第一检索词执行步骤 202a, 当执行完步骤 202a时， 图 3a即可变成图 3b。 参见图 3b, 图 3b为本发 明实施例提供的检索词之间的拓朴图结构第二示意图。 图 3b 示出了相 互连接的检索词之间的聚类关系， 其中， 实线连接的两个检索词表示该 两个检索词具有的聚类关系为： 两者被认为等价， 可聚类在一起； 虚线 连接的两个检索词具有的聚类关系为： 两者不等价， 不可聚类在一起， 后续可去掉该虚线。
由于在图 3a所示的拓朴图中，与同一个第一检索词相关的各个第二 检索词之间也可能具有潜在的聚类关系。 这种聚类关系可能已经在步骤 203找到 （比如， b3和 b4之间的聚类关系）， 也可能没有（比如， b2和 b3之间的聚类关系）。 为使检索词聚类更加精确， 依据图 2b所示的潜在 聚类关系挖掘流程， 可得到其中， 与用户提供的相关的之间潜在的聚类 关系可通过图 3c中的虚线表示的潜在的聚类关系。 以图 3c中用户提供 的第一检索词 bl 为例进行描述， 用户提供的其他检索词原理类似。 如 此， 根据上面图 3a的描述可以知道， 与 bl的第二检索词为： b2、 b3和 b4, 如此， 基于步骤 201b, 当 b2、 b3和 b4与 bl之间的相似度值均大 于或等于第二预设阈值时，本发明可以补充挖掘出 3条潜在的聚类关系： b2与 b3之间的聚类关系， b2与 b4之间的聚类关系， 以及 b3与 b4之 间的聚类关系。 其中， b3与 b4之间的聚类关系已经在上述步骤 202a被 确定， 因此， 作为本发明实施例的一种扩展， 本发明可省略执行确定 b3 与 b4之间的聚类关系的操作， 只需增加 b2与 b3之间的聚类关系和 b2 与 b4之间的聚类关系。 之后计算出 b2与 b3之间的相似度值， 以及 b2 与 b4之间的相似度值， 判断 b2与 b3之间的聚类关系和 b2与 b4之间 的聚类关系是否符合聚类的标准， 具体为： 基于上述步骤 202b, 判断该 b2与 b3之间的相似度值是否大于或等于第一预设阈值， 如果是， 则确 定 b2与 b3之间的聚类关系为： b2和 b3等价， 可聚类在一起， 否则， 确定 b2与 b3之间的聚类关系为： 不将 b2和 b3聚类在一起。 同理， b2 与 b4之间的相似度值也是执行类似方法。
当通过上面描述验证出图 3c中虚线连接的两个检索词等价，可以聚 类在一起时， 将该虚线变为实线； 否则， 保持该虚线不变， 即认为虚线 连接的两个检索词不等价， 不可聚类在一起， 后续可将该虚线去掉。 之 后， 将最终由实线连接的所有检索词作为本发明实施例最终的聚类结 果。
由于本发明实施例中， 检索词之间的聚类关系通过检索词之间的实 线（也称为边关系）表示， 因此， 本发明实施例可仅遍历边关系， 这样， 可以使本发明实施例的复杂度降低为 0(n+e),其中 n表示检索词的个数， e表示边关系的个数。
需要说明的是， 作为本发明实施例的一种扩展， 本发明实施例中， 还可进一步挖掘图 3a中与用户提供的第一检索词相关的第二检索词，以 及该第二检索词在 N (比如 N为 3 )跳以内各 "子孙" 节点之间潜在的 聚类关系， 具体实现方式参见图 2b所示流程， 这里不再详述。
另外， 在竟价广告系统中， 候选检索词集合并非固定不变， 其可随 着时间的推移递增检索词。 比如， 在某一个时间点， 候选检索词集合又 新增加了用户提供的第一检索词， 该新增加的第一检索词， 相对之前的 所有检索词是新出现的。 对该新增加的第一检索词， 也需要对其执行类 似图 2a和图 2b所示的聚类操作， 同时， 将执行聚类操作后得到的结果 与之前的聚类结果整合到一起。 具体见图 4所示的流程。
参见图 4, 图 4为本发明实施例提供的新增加第一检索词时的流程 (记为增量更新流程） 图。 如图 4所示， 该流程可包括以下步骤： 步骤 401 , 确定与该增加的第一检索词相关的第二检索词， 并将该 增加的第一检索词和该确定的与该增加的第一检索词相关的第二检索 词中与所述候选检索词集合中任一检索词不同的第二检索词添加到所 述候选检索词集合中。
比如， 候选检索词集合在执行步骤 401 之前存放的检索词为图 3a 所示的 bl至 b9, 当执行到本步骤 401时， 如果新增加以下两个第一检 索词： nl和 n2。 其中， 与 nl相关的第二检索词为 b5和 b6, 与 n2相关 的第二检索词为 bl、 b2、 b3、 b4、 b8和 n3 , 具体可参见图 3d所示。 由 于与 nl相关的 b5和 b6, 以及与 n2相关的 bl、 b2、 b3、 b4、 b8都已存 放在候选检索词集合中， 因此， 本步骤 401仅可将 nl、 n2, 以及与 n2 相关的 n3添加至候选检索词集合。
步骤 402, 根据检索词的文本特征和 /或语义特征对所述候选检索词 集合中新增加的第一检索词和与该第一检索词相关的第二检索词执行 聚类操作。
该聚类操作与图 2a所示的流程类似。下面仅以新增加的第一检索词 为 nl为例对本步骤 402进行描述， 增加的其他检索词原理类似。
贝' J , 针对 nl , 基于步骤 401 , 确定出与该 nl相关的第二检索词为 b5和 b6。 如此， 执行到本步骤 402时， 基于图 2a所示的流程， 则根据 nl的文本特征和 /或语义特征计算 nl与 b5之间的相似度值， 以及计算 nl与 b6之间的相似度值， 之后判断 nl与 b5之间的相似度值是否大于 或等于第一预设阈值， 如果是， 则确定 nl与 b5等价， 两者可以聚类在 一起， 否则， 不将 nl与 b5聚类在一起。 针对 nl与 b6之间的相似度值 也执行同样操作。
步骤 403 , 对候选检索词集合中与增加的第一检索词相关的第二检 索词进行潜在聚类关系的挖掘。 本步骤 403可采用图 2b所示的流程进行潜在聚类关系的挖掘， 筒 单描述为： 从候选检索词集合中与增加的第一检索词相关的各个第二检 索词， 或者从与增加的第一检索词聚类在一起的各个第二检索词中选取 与该第一检索词之间的相似度值大于或等于第二预设阈值的第二检索 词； 计算该选取的任意两个第二检索词之间的相似度值， 如果该计算的 相似度值大于或等于所述第一预设阈值， 则将该两个第二检索词聚类在 一起。
仍以新增加的第一检索词为检索词 nl为例， 由于在步骤 401确定 出与该 nl相关的第二检索词为 b5和 b6, 因此， 执行到本步骤 403时， 如果 b5和 b6分别与 nl之间的相似度值均大于第二预设阈值， 则可计 算 b5和 b6之间的相似度值， 如果该计算的相似度值大于或等于第一预 设阈值， 则将该两个检索词 b5和 b6聚类在一起， 否则， 不将 b5和 b6 聚类在一起。
至此，通过上述步骤 401至步骤 403实现了新增加的第一检索词（记 为增量检索词）与原来已存在的检索词 （记为旧检索词）之间的聚类关 系 （下文记为增量聚类结果）。 该增量聚类结果与之前存在的全量聚类 结果统称为本发明最终的聚类结果。
需要说明的是， 本实施例中， 与第一检索词相关的第二检索词并非 固定不变， 其也会根据用户增删检索词发生变化， 基于此， 本发明实施 例提供的方法还应能够反映该变化。 该变化通过定期更新候选检索词集 合（记为全量更新）实现， 具体实现为： 在设定的全量更新时间到达时， 针对所述候选检索词集合中的第一检索词， 确定与该第一检索词相关的 第二检索词， 将该第一检索词和确定的与该第一检索词相关的第二检索 词均放入一个新的候选检索词集合中，之后按照图 2a和图 2所示的流程 对该新的候选检索词集合中的检索词进行聚类， 得到全量聚类结果。 这 可通过表 1形象描述。
假设第一天用户提供的第一检索词为： B 该第一检索词相应的 QBM扩展结果为 Q^i B 中， 该扩展结果主要为与该第一检索词相关 的第二检索词的集合。 基于图 2a和图 2b所示的流程对第一检索词和第 二检索词进行聚类得到的聚类结果为： CF CXB ); 如此， 随着时间推
Figure imgf000013_0001
Figure imgf000014_0001
从表 1 可以看出， 全量更新在第 i天开始， 第 k天结束， 在第 k+1 (也即 L )天，做全量数据与增量数据的同步操作， 即， 将截止到第 k+1 (也即 L )天候选检索词集合中的所有第一检索词执行图 4所示的流程。 下面对本发明实施例提供的装置进行描述。
参见图 5 , 图 5为本发明实施例提供的装置的基本结构图。 如图 5 所示， 该装置可包括：
建立单元 501 , 用于建立候选检索词集合， 所述候选检索词集合包 含由用户提供的第一检索词、 以及与第一检索词相关的第二检索词； 聚类单元 502, 用于根据检索词的文本特征和 /或语义特征对所述候 选检索词集合中的第一检索词和与该第一检索词相关的第二检索词执 行聚类操作。
在具体实现时， 图 5所示的装置具体可参见图 6。
参见图 6, 图 6为本发明实施例提供的装置的详细结构图。 如图 6 所示， 该装置可包括建立单元 601和聚类单元 602, 其中， 建立单元 601 和聚类单元 602具有的功能分别与图 5所示的建立单元 501和聚类单元 502类似， 这里不再赘述。
优选地， 如图 6所示， 该装置可进一步包括：
添加单元 603 , 用于当用户增加新的第一检索词时， 确定与该增加 的第一检索词相关的第二检索词， 并将该增加的第一检索词和该确定的 与该增加的第一检索词相关的第二检索词中与所述候选检索词集合中 任一检索词不同的第二检索词添加到所述候选检索词集合中；
基于此，聚类单元 602还用于根据检索词的文本特征和 /或语义特征 对所述候选检索词集合中新增加的第一检索词和与该第一检索词相关 的第二检索词执行聚类操作。
优选地， 如图 6所示， 该装置进一步包括：
更新单元 604, 用于在设定的全量更新时间到达时， 针对所述候选 检索词集合中的第一检索词， 确定与该第一检索词相关的第二检索词， 将该第一检索词和确定的与该第一检索词相关的第二检索词均放入一 个新的候选检索词集合中。
基于此，聚类单元 602还用于根据检索词的文本特征和 /或语义特征 对该新的候选检索词集合中第一检索词和与该第一检索词相关的第二 检索词执行聚类操作。
具体地， 聚类单元 602通过以下子单元执行聚类操作：
计算子单元 6021 ,用于根据第一检索词的文本特征和 /或语义特征分 别计算该第一检索词和与该第一检索词相关的各个第二检索词之间的 相似度值；
聚类子单元 6022,用于在第一检索词和第二检索词之间的相似度值 大于或等于第一预设阈值时， 将该第一检索词和该第二检索词聚类在一 起。
优选地，聚类子单元 6022还用于从与第一检索词相关的各个第二检 索词， 或者从与第一检索词聚类在一起的各个第二检索词中选取与该第 一检索词之间的相似度值大于或等于第二预设阈值的第二检索词； 以及 计算该选取的任意两个第二检索词之间的相似度值， 如果该计算的相似 度值大于或等于所述第一预设阈值， 则将该两个第二检索词聚类在一 起， 所述第一预设阈值与第二预设阈值无关。
以上对本发明实施例提供的装置进行了描述。
由以上技术方案可以看出， 本发明提供的检索词的聚类方法和装置 在进行检索词聚类时， 并非像现有技术那样仅将用户提供的检索词进行 字面关系的聚类， 而是同时考虑到用户提供的检索词， 以及与该用户提 供的检索词相关的其他检索词，并根据检索词的文本特征和 /或语义特征 对用户提供的检索词， 以及与该用户提供的检索词相关的其他检索词进 行聚类， 这显然大大增加检索词聚类的准确度；
进一步地， 本发明还挖掘出与用户提供的第一检索词相关的各个第 二检索词之间的聚类关系， 这相比于现有技术， 可深度挖掘出检索词之 间的聚类关系， 使检索词的聚类更加精确。
以上所述仅为本发明的较佳实施例而已， 并不用以限制本发明， 凡 在本发明的精神和原则之内， 所做的任何修改、 等同替换、 改进等， 均 应包含在本发明保护的范围之内。

Claims

权利要求书
1、 一种检索词的聚类方法， 其特征在于， 该方法包括：
建立候选检索词集合， 所述候选检索词集合包含由用户提供的第一 检索词、 以及与第一检索词相关的第二检索词；
根据检索词的文本特征和 /或语义特征对所述候选检索词集合中的 第一检索词和与该第一检索词相关的第二检索词执行聚类操作。
2、根据权利要求 1所述的方法， 其特征在于， 当用户增加第一检索 词时， 该方法进一步包括：
确定与该增加的第一检索词相关的第二检索词， 并将该增加的第一 检索词和该确定的与该增加的第一检索词相关的第二检索词中与所述 候选检索词集合中任一检索词不同的第二检索词添加到所述候选检索 词集合中；
根据检索词的文本特征和 /或语义特征对所述候选检索词集合中新 增加的第一检索词和与该第一检索词相关的第二检索词执行聚类操作。
3、 根据权利要求 1所述的方法， 其特征在于， 该方法进一步包括： 在设定的全量更新时间到达时， 针对所述候选检索词集合中的第一 检索词， 确定与该第一检索词相关的第二检索词， 将该第一检索词和确 定的与该第一检索词相关的第二检索词均放入一个新的候选检索词集 合中，根据检索词的文本特征和 /或语义特征对该新的候选检索词集合中 第一检索词和与该第一检索词相关的第二检索词执行聚类操作。
4、根据权利要求 1至 3任一所述的方法， 其特征在于， 根据检索词 的文本特征和 /或语义特征对第一检索词和与该第一检索词相关的第二 检索词执行聚类操作包括：
根据第一检索词的文本特征和 /或语义特征分别计算该第一检索词 和与该第一检索词相关的各个第二检索词之间的相似度值， 如果第一检 索词和第二检索词之间的相似度值大于或等于第一预设阈值， 则将该第 一检索词和该第二检索词聚类在一起。
5、 根据权利要求 4所述的方法， 其特征在于， 该方法进一步包括： 从与第一检索词相关的各个第二检索词， 或者从与第一检索词聚类 在一起的各个第二检索词中选取与该第一检索词之间的相似度值大于 或等于第二预设阈值的第二检索词；
计算该选取的任意两个第二检索词之间的相似度值， 如果该计算的 相似度值大于或等于所述第一预设阈值， 则将该两个第二检索词聚类在 一起。
6、根据权利要求 1所述的方法， 其特征在于， 与第一检索词相关的 第二检索词包括：
与第一检索词匹配的检索词， 和 /或， 以第一检索词为关键词进行搜 索所得到的搜索结果中的检索词。
7、 一种检索词的聚类装置， 其特征在于， 该装置包括：
建立单元， 用于建立候选检索词集合， 所述候选检索词集合包含由 用户提供的第一检索词、 以及与第一检索词相关的第二检索词；
聚类单元，用于根据检索词的文本特征和 /或语义特征对所述候选检 索词集合中的第一检索词和与该第一检索词相关的第二检索词执行聚 类操作。
8、根据权利要求 7所述的装置，其特征在于，所述装置进一步包括： 添加单元， 用于当用户增加第一检索词时， 确定与该增加的第一检 索词相关的第二检索词， 并将该增加的第一检索词和该确定的与该增加 的第一检索词相关的第二检索词中与所述候选检索词集合中任一检索 词不同的第二检索词添加到所述候选检索词集合中； 所述聚类单元，还用于根据检索词的文本特征和 /或语义特征对所述 候选检索词集合中新增加的第一检索词和与该第一检索词相关的第二 检索词执行聚类操作。
9、根据权利要求 7所述的装置，其特征在于，所述装置进一步包括： 更新单元， 用于在设定的全量更新时间到达时， 针对所述候选检索 词集合中的第一检索词， 确定与该第一检索词相关的第二检索词， 将该 第一检索词和确定的与该第一检索词相关的第二检索词均放入一个新 的候选检索词集合中；
所述聚类单元还用于根据检索词的文本特征和 /或语义特征对该新 的候选检索词集合中第一检索词和与该第一检索词相关的第二检索词 执行聚类操作。
10、 根据权利要求 7至 9任一所述的装置， 其特征在于， 所述聚类 单元通过以下子单元执行聚类操作：
计算子单元，用于根据第一检索词的文本特征和 /或语义特征分别计 算该第一检索词和与该第一检索词相关的各个第二检索词之间的相似 度值；
聚类子单元， 用于在第一检索词和第二检索词之间的相似度值大于 或等于第一预设阈值时， 将该第一检索词和该第二检索词聚类在一起。
11、根据权利要求 10所述的装置， 其特征在于， 所述聚类子单元还 用于从与第一检索词相关的各个第二检索词， 或者从与第一检索词聚类 在一起的各个第二检索词中选取与该第一检索词之间的相似度值大于 或等于第二预设阈值的第二检索词； 以及计算该选取的任意两个第二检 索词之间的相似度值， 如果该计算的相似度值大于或等于所述第一预设 阈值， 则将该两个第二检索词聚类在一起。
PCT/CN2012/070824 2011-02-18 2012-02-01 检索词的聚类方法和装置 Ceased WO2012109959A1 (zh)

Priority Applications (1)

Application Number Priority Date Filing Date Title
US14/000,083 US20140019452A1 (en) 2011-02-18 2012-02-01 Method and apparatus for clustering search terms

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201110043030.7A CN102646103B (zh) 2011-02-18 2011-02-18 检索词的聚类方法和装置
CN201110043030.7 2011-02-18

Publications (1)

Publication Number Publication Date
WO2012109959A1 true WO2012109959A1 (zh) 2012-08-23

Family

ID=46658926

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2012/070824 Ceased WO2012109959A1 (zh) 2011-02-18 2012-02-01 检索词的聚类方法和装置

Country Status (3)

Country Link
US (1) US20140019452A1 (zh)
CN (1) CN102646103B (zh)
WO (1) WO2012109959A1 (zh)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103744889A (zh) * 2013-12-23 2014-04-23 百度在线网络技术（北京）有限公司 一种用于对问题进行聚类处理的方法与装置
WO2015016908A1 (en) 2013-07-30 2015-02-05 Intuit Inc. Method and system for clustering similar items

Families Citing this family (14)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103699550B (zh) * 2012-09-27 2017-12-12 腾讯科技（深圳）有限公司 数据挖掘系统及数据挖掘方法
CN103853722B (zh) * 2012-11-29 2017-09-22 腾讯科技（深圳）有限公司 一种基于检索串的关键词扩展方法、装置和系统
CN104123279B (zh) * 2013-04-24 2018-12-07 腾讯科技（深圳）有限公司 关键词的聚类方法和装置
CN104933081B (zh) 2014-03-21 2018-06-29 阿里巴巴集团控股有限公司 一种搜索建议提供方法及装置
TW201619853A (zh) * 2014-11-21 2016-06-01 財團法人資訊工業策進會 檢索過濾方法及其處理裝置
CN104462272B (zh) * 2014-11-25 2018-05-04 百度在线网络技术（北京）有限公司 搜索需求分析方法和装置
CN106326259A (zh) * 2015-06-26 2017-01-11 苏宁云商集团股份有限公司 搜索引擎中商品标签的构建方法、系统及搜索方法和系统
CN106610989B (zh) * 2015-10-22 2021-06-01 北京国双科技有限公司 搜索关键词聚类方法及装置
CN106951511A (zh) * 2017-03-17 2017-07-14 福建中金在线信息科技有限公司 一种文本聚类方法及装置
US11409799B2 (en) * 2017-12-13 2022-08-09 Roblox Corporation Recommendation of search suggestions
CN111259058B (zh) * 2020-01-16 2023-09-15 北京百度网讯科技有限公司 数据挖掘方法、数据挖掘装置和电子设备
CN112650907B (zh) * 2020-12-25 2023-07-14 百度在线网络技术（北京）有限公司 搜索词的推荐方法、目标模型的训练方法、装置及设备
CN112905765B (zh) * 2021-02-09 2024-06-18 联想(北京)有限公司 一种信息处理方法及装置
CN115376054B (zh) * 2022-10-26 2023-03-24 浪潮电子信息产业股份有限公司 一种目标检测方法、装置、设备及存储介质

Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20100131563A1 (en) * 2008-11-25 2010-05-27 Hongfeng Yin System and methods for automatic clustering of ranked and categorized search objects
KR20100106718A (ko) * 2009-03-24 2010-10-04 엔에이치엔(주) 연관 키워드에 따른 클러스터를 이용하여 검색 키워드를 분류하는 시스템 및 방법

Family Cites Families (19)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US5488725A (en) * 1991-10-08 1996-01-30 West Publishing Company System of document representation retrieval by successive iterated probability sampling
US5931907A (en) * 1996-01-23 1999-08-03 British Telecommunications Public Limited Company Software agent for comparing locally accessible keywords with meta-information and having pointers associated with distributed information
US6502091B1 (en) * 2000-02-23 2002-12-31 Hewlett-Packard Company Apparatus and method for discovering context groups and document categories by mining usage logs
EP1182581B1 (en) * 2000-08-18 2005-01-26 Exalead Searching tool and process for unified search using categories and keywords
KR20020049164A (ko) * 2000-12-19 2002-06-26 오길록 유전자 알고리즘을 이용한 카테고리 학습과 단어클러스터에 의한 문서 자동 분류 시스템 및 그 방법
US20030120630A1 (en) * 2001-12-20 2003-06-26 Daniel Tunkelang Method and system for similarity search and clustering
US6947930B2 (en) * 2003-03-21 2005-09-20 Overture Services, Inc. Systems and methods for interactive search query refinement
US7428529B2 (en) * 2004-04-15 2008-09-23 Microsoft Corporation Term suggestion for multi-sense query
US7260568B2 (en) * 2004-04-15 2007-08-21 Microsoft Corporation Verifying relevance between keywords and web site contents
US7689585B2 (en) * 2004-04-15 2010-03-30 Microsoft Corporation Reinforced clustering of multi-type data objects for search term suggestion
US7756855B2 (en) * 2006-10-11 2010-07-13 Collarity, Inc. Search phrase refinement by search term replacement
US7792858B2 (en) * 2005-12-21 2010-09-07 Ebay Inc. Computer-implemented method and system for combining keywords into logical clusters that share similar behavior with respect to a considered dimension
US8799285B1 (en) * 2007-08-02 2014-08-05 Google Inc. Automatic advertising campaign structure suggestion
US7962486B2 (en) * 2008-01-10 2011-06-14 International Business Machines Corporation Method and system for discovery and modification of data cluster and synonyms
US20100094673A1 (en) * 2008-10-14 2010-04-15 Ebay Inc. Computer-implemented method and system for keyword bidding
US8463783B1 (en) * 2009-07-06 2013-06-11 Google Inc. Advertisement selection data clustering
US9002857B2 (en) * 2009-08-13 2015-04-07 Charite-Universitatsmedizin Berlin Methods for searching with semantic similarity scores in one or more ontologies
US20110295678A1 (en) * 2010-05-28 2011-12-01 Google Inc. Expanding Ad Group Themes Using Aggregated Sequential Search Queries
US9830379B2 (en) * 2010-11-29 2017-11-28 Google Inc. Name disambiguation using context terms

Patent Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20100131563A1 (en) * 2008-11-25 2010-05-27 Hongfeng Yin System and methods for automatic clustering of ranked and categorized search objects
KR20100106718A (ko) * 2009-03-24 2010-10-04 엔에이치엔(주) 연관 키워드에 따른 클러스터를 이용하여 검색 키워드를 분류하는 시스템 및 방법

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2015016908A1 (en) 2013-07-30 2015-02-05 Intuit Inc. Method and system for clustering similar items
EP3031024A4 (en) * 2013-07-30 2017-01-11 Intuit Inc. Method and system for clustering similar items
CN103744889A (zh) * 2013-12-23 2014-04-23 百度在线网络技术（北京）有限公司 一种用于对问题进行聚类处理的方法与装置

Also Published As

Publication number Publication date
CN102646103B (zh) 2016-03-16
CN102646103A (zh) 2012-08-22
US20140019452A1 (en) 2014-01-16

Similar Documents

Publication Publication Date Title
WO2012109959A1 (zh) 检索词的聚类方法和装置
CN103187052B (zh) 一种建立用于语音识别的语言模型的方法及装置
CN109523991B (zh) 语音识别的方法及装置、设备
US10698654B2 (en) Ranking and boosting relevant distributable digital assistant operations
EP2518978A2 (en) Context-Aware Mobile Search Based on User Activities
US11264006B2 (en) Voice synthesis method, device and apparatus, as well as non-volatile storage medium
CN105224554A (zh) 推荐搜索词进行搜索的方法、系统、服务器和智能终端
CN103699689A (zh) 事件知识库的构建方法及装置
CN110019348A (zh) 扩展搜索查询
CN105900081A (zh) 基于自然语言处理的搜索
CN104216906A (zh) 语音搜索方法和设备
CN103383699A (zh) 字符串检索方法及系统
CN106777328B (zh) 一种移动终端的题目推荐方法及装置
WO2017054332A1 (zh) 路径查询方法、装置、设备及非易失性计算机存储介质
CN102955821A (zh) 一种对查询序列进行扩展处理的方法与设备
CN104036018A (zh) 视频获取方法和装置
CN105930376A (zh) 一种搜索方法和装置
CN110232129A (zh) 场景纠错方法、装置、设备和存储介质
CN102063194A (zh) 用于供用户进行文字输入的方法、设备、服务器和系统
CN103970756A (zh) 热点话题提取方法、装置和服务器
CN103092928A (zh) 语音查询方法及系统
CN110246493A (zh) 通讯录联系人查找方法、装置及存储介质
US20170228402A1 (en) Inconsistency Detection And Correction System
CN106682190A (zh) 标签知识库的构建方法、装置、应用搜索方法和服务器
CN103064825B (zh) 模糊音对建立、设置方法和输入法及其装置和系统

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 12747612

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

WWE Wipo information: entry into national phase

Ref document number: 14000083

Country of ref document: US

32PN Ep: public notification in the ep bulletin as address of the adressee cannot be established

Free format text: NOTING OF LOSS OF RIGHTS PURSUANT TO RULE 112(1) EPC (EPO FORM 1205N DATED 15/01/2014)

122 Ep: pct application non-entry in european phase

Ref document number: 12747612

Country of ref document: EP

Kind code of ref document: A1