Patent Application
20090259642
There are many different types of techniques for discovering information, using a computer network. One specific technique is referred to as a community-based question and answering service (referred to as cQA). The cQA service is a kind of web service through which people can post questions and also answer other people's questions on a web site. The growth of cQA has been relatively significant, and it has recently been offered by commercially available web search engines.
In current cQA services, a community of users either subscribes to the service, or simply accesses the service through a network. The users in the community can then post questions that are viewable by other users in the community. The community users can also post answers to questions that were previously submitted by other users. Therefore, over time, cQA services build up very large archives of previous questions and answers posted for those previous questions.
In some current cQA systems, once the user who has asked a question and has received a sufficient number of answers, the user can then choose the best answer from those answers that have been received. The user may also simply leave it to the community to vote for the best answer. This results in large question and answer archives, representing knowledge repositories, that can be used later.
For instance, these repositories can be used to expedite answers to the user. In other words, there is typically a lag time between the time when a user in the community posts a question, and the time when other users of the community post answers to that question. In order to avoid this lag time, some cQA services automatically search the archive of questions and answers to see if the same question has previously been asked. If the question is found in the archives, then one or more previous answers can be provided, in answer to the current question, with very little delay. This type of searching for previous answers is sometimes referred to as “question search” or “on-line search”.
The discussion above is merely provided for general background information and is not intended to be used as an aid in determining the scope of the claimed subject matter.
In a question answering system, the system identifies a type of question input by a user. The system then generates answer summaries that summarize answers to the input question in a format that is determined based on the type of question asked by the user. The answer summaries are output, in the corresponding format, in answer to the input question.
This Summary is provided to introduce a selection of concepts in a simplified form that are further described below in the Detailed Description. This Summary is not intended to identify key features or essential features of the claimed subject matter, nor is it intended to be used as an aid in determining the scope of the claimed subject matter. The claimed subject matter is not limited to implementations that solve any or all disadvantages noted in the background.
Answer summarization component 106 then generates answer summaries, in a form corresponding to the type of the question identified. This is indicated by block 124 in
In analyzing the types of questions asked in cQA services, it has been found that some types of questions have best answers that are reusable in later searching, while others do not. For instance, questions that ask for answers that are facts may likely have a best answer that is extremely useful in later searching. However, cQA questions that ask for an opinion may not be helpful in later information searching.
For example, a fact-based, closed ended question, such as “What was the birth date of Abraham Lincoln?” will have one unique answer. Therefore, an answer that is labeled in a cQA system as the best answer to that question will likely be reusable and very relevant in later information searching. However, a question such as “What actress has the most seductive voice?” asks for an opinion. Therefore, even the answer labeled by the community as the best answer is still only an opinion. Thus, those answers may not be as reusable in later information searching processes.
Similarly, even fact-based open-ended questions may not be as reusable as a closed-ended question. For instance, a question such as “Does anyone know of any movie stars having birthdays coming up soon?” is open ended. First of all, the time period of the question is relevant because it asks for birthdays that are “coming up soon”. Similarly, there may be a wide variety of different people answering the question, and so they may provided a wide variety of different birthdays for different movie stars, depending on the type of movie that they enjoy. Thus, answers to this type of fact-based open-ended question, even if they are labeled as “best answers” may be less reusable or less relevant to later information searching processes.
Therefore, in order to generate summaries of answers to cQA questions, question type identifier 104 identifies the question type, and answer summarization component 106 identifies the types of answers to the identified question, and generates a summary of the best answers (and optionally other answers as well) in a format that is determined based on the type of the question identified.
The reusable best answers are further divided into factual best answers 208 and subjective best answers 210. A factual best answer is a fact that can be used as the best answer, while a subjective best answer is an opinion type of answer that can be used as the best answer.
The factual best answers 208 can be divided into two further types, those being unique best answers 212 and non-unique best answers 214. A unique best answer has only a single best answer to its question. A non-unique best answer has alternative best answers.
These alternative (non-unique) best answers 214 can be divided into direct answers 216 and indirect best answers 218. A direct answer is just that, an answer that answers its question directly. An indirect answer is an answer which answers the question through inference.
The non-reusable best answers 206 are divided into two types: relevant best answers 220 and irrelevant best answers 222. Relevant best answers 220 are not reusable as a best answer to its question, if that question is asked at a later time, but they are relevant to the question. An irrelevant best answer is not reusable as a best answer to its question and it is irrelevant to its question.
Table 1 summarizes answer type taxonomy 200 and gives examples for each answer type.
Table 1 first shows a reusable, factual, unique best answer. This best answer type provides enough useful information in the answer and no other answers add more information. For instance, if a question asks “Which country has the largest population?” and the best answer is “China”. This answer is accurate and complete, so it can be categorized as the unique type.
The difference between direct and indirect answer types is a bit subtle. A direct best answer answers its question in a straightforward manner; while an Indirect best answer answers its question in an inferred manner. Answerers of Indirect best answers usually look beyond the surface level of the question and try to give answers that satisfy the inferred needs behind question. For example, a question asked “Where can i download fatwire from because every website i go on wants to charge me?” and the best answer was “http://www.duckwire.com/”. This answer addresses the ‘free’ and ‘download’ parts of the question. However, instead of giving a FatWire download site, the BA suggested an alternative, i.e. DuckWire, that presumably would fulfill what the asker really wanted. Notice that a possible Direct BA suggested that download the basic version of FatWire from Fatwire.com that is free. Assume FatWire is a popular peer to peer (P2P) file sharing tool, and that it has two versions. The Basic version is free but the Pro version is not free. It can be inferred from the question that the asker wanted the Pro version and not the Basic. A true Direct BA is a site where the asker could download the Pro version without paying but it may very likely be an illegal site even if it does exist. Therefore the DuckWire suggestion is likely a very good alternative. Another example of Direct vs. Indirect BA is shown earlier in Table 1. The question asked celebrity birthdays in the coming month (November). A Direct BA contains a list of celebrity birthdays in November as shown in one of the answers in Table 1. A better alternative, i.e. an Indirect one, showed the asker where she can find this information herself not only for November but for all other months.
A Subjective BA answers questions that look for personal opinions or recommendations. For example, a question asked “What is the best sci-fi movie u ever saw?” Each answerer would have his or her own opinion about what the best sci-fi movie is. There might be consensus but it will likely not be unique. Many answers in the Entertainment & Music category are this type.
Answers that fall into the Not Reusable category could be relevant or irrelevant to their questions. Despite their relevance, they could not be used as the BAs. For example, a question asked “Why was “The ABC Song” by John Doe so shortlived?” A Relevant BA said “I'm not sure where you live, but in NJ, especially South Jersey, that song was played out. It may have made the New Year's “We're never playing this song again” list.” This BA seems relevant to the question but we could not be sure it is the BA without knowing where the asker resided.
There are many possible reasons that make a BA irrelevant. Two exemplary possible explanations are listed in Table 1. One is that the BA does not answer its question. For example, a question asked for how to forward an email without showing the email addresses in the To box. There is an answer that contains the correct method “put the email address into the bcc box”, but the best answer voted by one community is: “It appears that the question period has expired. If an answer has been given that meets your needs, please pick a ‘best answer’.” Another one is that the BA's question is nonsense and therefore its answers are as well. For example, someone asked “*oink* Whose gunna partay with me?” Though a BA was chosen, it does not make sense to reuse the BA.
The informational question 306 includes questions that request information about topics, and can be divided into constant questions 312 and dynamic questions 314. Constant questions have a fixed or stable set of answers, while dynamic questions do not.
The dynamic questions 314 can be further divided into three categories, those including opinion 316, context dependent questions 318 and open questions 320. Opinion questions are those asking for opinions. Questions in this category seek opinions from people in the cQA community about what they think of someone, some event, some object, etc. Context-dependent questions 318 are those questions which have different answers, depending on the different context in which they are asked. For example, the question “What is the population of China?” will have different answers according to the date on which it is asked. Open questions 320 are those questions that ask for some facts or methods but have a variety of answers, or the answer itself may have unconstrained depth.
The transactional questions 308 can further be divided into possessive questions 322 and interactive questions 324. Possessive questions 322 indicate the askers intention to own the resource requests, regardless of whether a computer is required to make use of the required resource. The interactive questions 324 are questions looking for entertainment or utility resources.
Social questions 310 do not intend to obtain any given answer. These questions include telling jokes, and expressing the askers own ideas. For instance, one social question may be “Why do so many lazy people come on here simply just to ask . . . ?” The question is actually expressing a negative sentiment toward the large number of people who ask how to accomplish a certain task in the cQA community.
Table 2 illustrates question taxonomy 300, and gives examples of each type of question.
Identifying the question type and answer type as set out in
In using these taxonomies and analyzing large archives of cQA responses, it has been found that a number of the different question types have answers, less than 50 percent of which can be reused in later searching given a similar type of question. Therefore, in one embodiment, answer summarization component 106 uses question-focused multi-document summarization techniques to generate the best answers for reuse in answering questions that are submitted again, later, in the cQA service.
Generating summaries is discussed in greater detail below with respect to Tables 3-8. In any case, however, once the summaries are generated, based on the type of question that the answers are provided for, the questions and answer summaries (which may be best answer summaries) are output for use in other information searching systems (such as in question answer systems). In that instance, the questions with summarized answers can be stored in data store 110 shown in
Some illustrative embodiments of summarizing answers, based on question type, will now be described. In doing this, answer summarizer component 106 illustratively invokes one of a plurality of different summarization modules 1-N, one of which exists for each different question type. Examines of some illustrative summarization modules are shown in tables 3, 5, and 6. The present discussion proceeds with respect to applying question-focused multi-document summarization techniques to create answer summaries for open and opinion questions (questions 316 and 320) as shown in
In one embodiment, answer summarization component 106 employs the summarization steps set out in Table 3 in order to summarize open-type questions 320. The steps in Table 3 follow multi-document summarization procedures which include topic identification, interpretation and fusion, and then summary generation. In the topic identification phase, answers are clustered using any of a wide variety of relatively simple clustering techniques. For instance, cosine measure, which is a known similarity measure, can be used to estimate answer similarity and an empirically determined threshold can be used to merge answers into clusters. This is indicated in step 1 of Table 3.
Next, in order to understand the main idea of each cluster, salient noun phrases can be extracted using any of a wide variety of shallow parsing components. This is indicated in step 2 of Table 3. Each label (or noun phrase) in each cluster can then have a relevance score calculated for it, such as by using a known first-order relevance scoring function also sometimes referred to as automatic labeling of multi-nomial topic models. This is set out in step 3 in Table 3. The relevance function can be used to extract a key answer that contains the noun phrase that has the highest score (calculated in step 3) in each cluster, and the key answers can also be ranked, within each cluster, based on the relevance score. This is indicated by step 4 in Table 3.
Finally, the summarized answer can be presented to the user, by extracting the most important answer in each cluster, and sorting those most important answers according to the cluster size that spawned the answers. This is indicated by step 5 in Table 3.
Table 4 illustrates an example of summarization of results of an open-type question. The question asked methods of changing a first style of operating system to a second style. There are many kinds of software that have this functionality. The best answer to the question only listed a single choice-The ACMEDOCK product while other answers suggested ABC and DEF and GHI products. The summarized answer contains a variety of software that have this functionality and a web page link that lists all kinds of software for turning the first style of operating system into the second style.
Table 5 shows a question-focused answer summarization technique for opinion type questions. Opinion type questions can be divided into two subcategories. One is sentiment-oriented questions that ask the sentiment about something. For instance, “What do you think of . . . ” is a sentiment type question. The other type of opinion question is a list-oriented question that requests a list of answers and looks to determine which item is most popular. In sentiment-oriented questions, the askers likely care about how many people support or oppose something. Table 5 shows steps for summarizing answers to sentiment-type questions. Table 6 shows steps for summarizing list-type questions. Tables 7 and 8 show summaries generated for sentiment-type questions and list-type questions, respectively. Of course, the methods in Tables 5 and 6 and the summaries in Tables 7 and 8 are exemplary only.
As shown in Table 5, Answers are summarized by first determining whether they begin with a negative queue phrase or a positive queue phrase. This can be done simply by using an opinion-word dictionary that is a dictionary of opinion words, and a queue phrase list that lists queue phrases into positive and negative queue phrases. If the answer beings with a negative queue phrase, it is annotated as being “against” what is proposed, and if it begins with a positive queue phrase it is annotated as “supporting” what is proposed.
If the answer cannot be annotated in step 1, then, for a clause, answer summarizer 106 determines whether a number of positive sentiment words in the clause is larger than the number negative sentiment words. If so, the clause is annotated “positive” and if not, it is annotated “negative”.
In step 3, if there are negative indicators such as “don't”, “never”, etc. in front of the clause, then the sentiment is reversed.
In step 4, if the number of negative clauses is larger than the number of positive clauses in the answer, then the sentiment of the answer is “negative”. Otherwise, it is “positive”.
Finally, in step 5 of Table 5, the sentiment value of the question (is denoted as S(Q)), and the sentiment value of the answer is denoted as (S(A)). The final sentiment of the answer is the logical Boolean AND of S(Q) and S(A).
In Table 6, the steps for summarizing answers to list questions include first segmenting the answers into sentences and clustering the sentences using the clustering technique set out above in Table 3. This is indicated by steps 1 and 2 in Table 6.
Next, for each cluster, a key sentence based on mutual information between itself and other sentences within the cluster is chosen. The key sentences are then ranked by cluster size that spawned the key sentences. This is indicated by steps 3 and 4 in Table 6.
For the list-oriented questions, it can be seen that a simple counting algorithm tallies different answers of questions, together with their supporting votes, to obtain the answer summaries.
It can be seen that Table 7 presents the summarization result of a sentiment-oriented question which asks whether it is strange for a 16 year old child to talk to a teddy bear. The best answer is a negative response. However, if all answers are considered, it can be seen that half of the answers agree, but another half of them disagree. This tally is thus presented in the answer summarization.
Table 8 shows the summarization result of a list-oriented question which asks for users to identify the best science fiction movie they have ever seen. The best answer just gives a single choice, while the summarized answer gives a list of best science fiction movies with the number of supporting votes.
Question search component 354 first receives an input question 352 from a user. This is indicated by block 380 in
In any case, once the similar questions are identified in data store 110, the answers to those questions are retrieved from data store 110. This is indicated by block 384 in
It should also be noted that question search component 354 can be any type of information search component, such as an information retrieval system. In that type of system, data store 110 may be a valuable repository of knowledge that can be searched in response to an information retrieval query. Therefore, system 354 is not limited to a community question answering system, but can be any other type of information seeking system as well, that accesses the data stored in store 110 based on community based question answering services.
With reference to
Computer 510 typically includes a variety of computer readable media. Computer readable media can be any available media that can be accessed by computer 510 and includes both volatile and nonvolatile media, removable and non-removable media. By way of example, and not limitation, computer readable media may comprise computer storage media and communication media. Computer storage media includes both volatile and nonvolatile, removable and non-removable media implemented in any method or technology for storage of information such as computer readable instructions, data structures, program modules or other data. Computer storage media includes, but is not limited to, RAM, ROM, EEPROM, flash memory or other memory technology, CD-ROM, digital versatile disks (DVD) or other optical disk storage, magnetic cassettes, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other medium which can be used to store the desired information and which can be accessed by computer 510. Communication media typically embodies computer readable instructions, data structures, program modules or other data in a modulated data signal such as a carrier wave or other transport mechanism and includes any information delivery media. The term “modulated data signal” means a signal that has one or more of its characteristics set or changed in such a manner as to encode information in the signal. By way of example, and not limitation, communication media includes wired media such as a wired network or direct-wired connection, and wireless media such as acoustic, RF, infrared and other wireless media. Combinations of any of the above should also be included within the scope of computer readable media.
The system memory 530 includes computer storage media in the form of volatile and/or nonvolatile memory such as read only memory (ROM) 531 and random access memory (RAM) 532. A basic input/output system 533 (BIOS), containing the basic routines that help to transfer information between elements within computer 510, such as during start-up, is typically stored in ROM 531. RAM 532 typically contains data and/or program modules that are immediately accessible to and/or presently being operated on by processing unit 520. By way of example, and not limitation,
The computer 510 may also include other removable/non-removable volatile/nonvolatile computer storage media. By way of example only,
The drives and their associated computer storage media discussed above and illustrated in
A user may enter commands and information into the computer 510 through input devices such as a keyboard 562, a microphone 563, and a pointing device 561, such as a mouse, trackball or touch pad. Other input devices (not shown) may include a joystick, game pad, satellite dish, scanner, or the like. These and other input devices are often connected to the processing unit 520 through a user input interface 560 that is coupled to the system bus, but may be connected by other interface and bus structures, such as a parallel port, game port or a universal serial bus (USB). A monitor 591 or other type of display device is also connected to the system bus 521 via an interface, such as a video interface 590. In addition to the monitor, computers may also include other peripheral output devices such as speakers 597 and printer 596, which may be connected through an output peripheral interface 595.
The computer 510 is operated in a networked environment using logical connections to one or more remote computers, such as a remote computer 580. The remote computer 580 may be a personal computer, a hand-held device, a server, a router, a network PC, a peer device or other common network node, and typically includes many or all of the elements described above relative to the computer 510. The logical connections depicted in
When used in a LAN networking environment, the computer 510 is connected to the LAN 571 through a network interface or adapter 570. When used in a WAN networking environment, the computer 510 typically includes a modem 572 or other means for establishing communications over the WAN 573, such as the Internet. The modem 572, which may be internal or external, may be connected to the system bus 521 via the user input interface 560, or other appropriate mechanism. In a networked environment, program modules depicted relative to the computer 510, or portions thereof, may be stored in the remote memory storage device. By way of example, and not limitation,
Although the subject matter has been described in language specific to structural features and/or methodological acts, it is to be understood that the subject matter defined in the appended claims is not necessarily limited to the specific features or acts described above. Rather, the specific features and acts described above are disclosed as example forms of implementing the claims.
