Patent Application
20080183681
The present invention relates to providing relevant information to users, and in particular to providing relevant information to users with reduced user input.
The Internet has become a popular source of entertainment and information. Most Internet content is designed for access via a web browser, making it difficult for access via most consumer electronics (CE) devices which lack typical computer keyboards. As a result, the Internet is generally restricted to access on personal computers (PC) or via cumbersome interfaces on CE devices.
With advances in hardware and software technologies, CE devices are becoming more powerful. Growth in network infrastructure and the falling prices of hardware have increased the availability of network-capable entertainment devices. Many users are configuring home networks including cable set-top boxes, digital television sets, home media servers, digital audio players, personal video recorders, etc. Home network consumers are also creating, storing and accessing more digital content through CE devices and PCs.
A second trend, running in parallel to the emergence of networked entertainment devices, is the growing use of the Internet for creating and publishing content. Greater broadband penetration and falling memory prices are enabling users to move ever larger media files, such as television (TV) shows and full-length movies, through the Internet.
However, there is a gap between the digital content on the Internet and the networked digital entertainment devices in that most Internet content is structured and organized for access via a web browser not a typical CE device. For example, typically a user searches for Internet information using a search engine or by directly accessing a known website via a PC. When using a search engine, the user is required to form an initial query and then iteratively refine the query depending upon the results obtained. As such, the user is forced to comprehend and analyze large quantities of information to identify/access the exact information the user is looking for. This process may work on a PC, but on CE devices that lack a keyboard and a mouse, the searching/refinement process is awkward and unpleasant. Moreover, users typically expect a “lean-back” experience when it comes to using CE devices in their homes and home networks. For instance, someone watching a television news program on a television may not be inclined to conduct an Internet search if such a search requires any effort more than pushing a few buttons on a remote control.
The present invention provides a method and system for facilitating information searching for a user of an electronic device. One embodiment involves forming a query to search for information related to the user activity on the electronic device; resolving the query by searching available sources including one or more external sources for said related information; receiving an event indicating availability of related information; and providing the related information to the user.
Resolving the query may further include searching said available sources, and providing extracted related information to the user. Further, receiving an event indicating availability of the related information may further include receiving an event indicating availability of additional related information. In addition, providing the related information to the user may further include providing said additional related information to the user.
These and other features, aspects and advantages of the present invention will become understood with reference to the following description, appended claims and accompanying figures.
The present invention provides a method and system for facilitating access to information via electronic devices such as consumer electronic (CE) devices. One embodiment involves enabling home users to easily find and access Internet content related to content presented on a CE device. An example is enabling a user to easily find and access Internet content related to a program the user is watching on a television. The user is now able to access relevant information and video content on the Internet in a “lean-back” living room experience while watching TV.
Searching for information on the Internet typically involves two stages: search query formation, and data search and analysis. Query information involves forming a search query that describes the type of information being sought. Data search and analysis involves resolving the search query according to the following steps: potential sources of data are identified; relevant data from such sources are extracted via search queries and then aggregated (collected); and correlations in the form of associations among the aggregated data are identified to make the results more meaningful.
An example implementation for CE devices in a local area network (LAN), such as a home network, is described below, however the present invention is useful with other electronic devices, and electronic devices that are not in a LAN but have access to the Internet.
The network 10 further includes a search facilitator system 24 that provides searching, aggregation and analysis functions. The facilitator 24 performs query formation, data search and analysis, wherein query formation includes identifying potential search queries (i.e., potential data of interest to the user) based on the user's context. Further, data search and analysis includes extracting, aggregating and correlating data of interest using execution plans.
Information of interest to the user, or user-related information, may include one or more of: user profile, user preferences, content previously/currently accessed by the user, terms previously selected by the user, etc.
In one example, the client module 64 enables the user to obtain desired information from, e.g., the Internet using a simple and intuitive Graphical User Interface (GUI) application, utilizing the facilitator 24, including:
The user utilizes the client module 64 to access certain content, and the facilitator 24 obtains information related to the accessed content for display to the user. The user then requests that the facilitator 24 provide more information about the accessed content. For example, the user utilizes the client module 64 to request that the facilitator 24 provide more information from Internet data sources 66 about a pre-recorded/broadcast TV program the user is watching on the DTV 30.
Using the client module 64, the user can choose, edit or enter new queries (such as the suggested keywords/categories) with minimal effort on a CE device that may not have a keyboard/mouse. Specifically, the facilitator 24 suggests and displays queries including keywords related to the TV program and information categories related to those keywords. Using the suggested keywords and categories as search queries, users can seamlessly browse/search for related information available on the Internet through their CE devices by simply selecting among the suggested queries for searching. The facilitator 24 identifies many relevant search queries, and allows the user to edit a suggested query or enter a new query. The facilitator 24 then obtains information of interest to the user and presents such information to the user.
In the architecture shown in
In one example, the query identification function 25 identifies potential data of interest to the user, based on the user's current application state. Current application state refers to the state of the application that the user is using at the time the user desires to access relevant Internet content. For example, if the user is watching a television program on DTV 30, the channel the DTV is tuned to and the program being broadcast, constitute the application state.
The query identification function 25 identifies the content used/rendered by the application. Then, the query identification function 25 obtains metadata information and/or other associated data for the content being accessed, and identifies potential search queries that might represent the data of interest to the user. When a user accesses content that has structured meta-data available, the query identification function 25 directly uses field/value pairs from the metadata as potential search queries. For example, if a user is listening to a music album by the artist “Sting” and expresses interest to access related content, the query identification function 25 obtains the following fields from the album's metadata (content=“MusicAlbum” & artist=“Sting”) and using these, the query identification function 25 infers that the user might be interested to access more albums by the same artist and suggests (MusicAlbum, artist, “Sting”) as one of the search queries to the user.
When a user accesses content such as broadcast TV programs and DVDs, the query identification function 25 uses the caption data (closed captions), that is embedded in the content stream, to identify potential search queries. This embedded caption data contains useful information in the form of keywords. When a user watches a TV program and expresses interest to access related content, the query identification function 25 analyzes the TV program's caption text to identify significant keywords and suggests them to the user as possible search queries.
The query identification function 25 can be implemented, e.g., in a stand-alone module, in a device 20 such as a set-top box or in a CE device 30 such as a DTV. A user interface (UI) can be displayed on a device in the network/system 10 capable of displaying information, such as a CE device 30. An example of identifying keywords and suggesting them as possible search keywords by the query identification function 25 utilizing natural language processing (NLP) to analyze closed captions and identify keywords from the captions, is described below.
The closed captions (CC) of a TV program are embedded in the TV signal by the content provider before it is broadcast. They are primarily for the benefit of the hearing impaired. Extracting useful information from this text is not straightforward. The captions typically do not contain any case information, precluding any attempt to extract proper nouns based on case information. Also, they are often ungrammatical (e.g., because of the spoken nature of the content), poorly punctuated and may have typos. Because of these limitations, typical keyword extraction techniques used for text documents may not be suitable for closed caption text. In addition, the content of closed captions is highly dependent on the type of the program. A news program's captions are high-content and factual, whereas a sitcom's captions are typically low on content and full of slang.
The CCA 70 operates in real-time on broadcast signals and processes a steady stream of closed caption text 74 entering the system. The CCA maintains two history windows over the stream of incoming text (
A CC Tokenizer 78 receives the stream of CC text 74 and breaks it down into sentences. This is done in order to preserve the grammar of the text. A tagger 73 then tags sentences, e.g., using Brill's part-of-speech tagging. The tagger 73 analyzes the sentence and determines how each word is used in the sentence. The tagger 73 uses lexical rules to assign an initial tag to each word in a sentence, and then uses contextual rules to update the tag based on the context in which the word occurs. The contextual rules are sensitive to the grammar of the input sentence. Ungrammatical or incomplete sentences may result in incorrect tagging of the words in the sentence.
In one example, for an input sentence: “John Wayne ran home”:
The output of tagger 73 would be:
This indicates that in the previous sentence, “John” and “Wayne” are used as proper nouns, “ran” is a verb in past tense and “home” is a noun.
This tagged sentence from the tagger 73 is then passed on to a rule engine 79 which extracts keywords from the tagged sentence based on extraction policy rules from a rule library 71. A rule library 71, R, is an exhaustive set of rules that can be used to extract different kinds of phrases appearing in the sentence. The rules are represented as tag patterns. For example, it may have a rule to extract consecutive proper nouns (<PROP>+) and another rule to extract an adjective followed by one or more nouns (<ADJ><NOUN>+), etc. A rule selector 72 includes a mapping from genre to an extraction policy. The genre of the program being watched determines the type of keywords to extract from the captions. For example, if the program being watched is a high-content, factual program such as news, the extraction policy is highly aggressive, essentially extracting additional differing types of keywords (e.g., sequences of nouns, compound nouns, proper nouns etc.). On the other hand, if the program is a low-content, non-factual program such as a sitcom, a very conservative extraction policy is used, extracting keywords very selectively, extracting only those keywords considered as having a higher likelihood of being useful (e.g., only proper nouns). The rule engine 79 alters its extraction behavior depending upon the type of program being watched.
Each extraction policy, Pe, corresponds to a subset of the rules in R. This mapping can either be preset, or it can be learned. The mapping essentially defines the kinds of patterns to be used for extracting keywords 76 from a particular type (genre) of program. In one example, the mapping can be determined by conducting a small user study involving four subjects asked to mark the keywords they would like to search for from CC transcripts of four types of sample programs: News, Sitcom, Talk Show and Reality TV. The transcripts were then tagged using Brill's tagger and the tags of the marked keywords were extracted as rules (e.g., if the keyword “Global Warming” in a news program was marked, and if the words were tagged “Global<ADJ> Warming<NOUN>”, then “<ADJ><NOUN>” is extracted as a rule for the genre “news”). The top ranking rules (based on frequency and a threshold) were used as the rules that form the extraction policy for that kind of program and the union of all rules for all types of programs forms R. This facilitates reusability of rules and extraction policies. The rule engine 79 applies the extraction policy on the text received from the tagger 73 and extracts keywords from it. These keywords are then weighted based on whether they occur in the most recent window. The weighted keywords are then ordered and presented to the user.
The extracted keywords identify information of potential interest to the user. The query resolution function 27 enables extracting data related to identified data of potential interest to the user, aggregating the extracted data and correlating the aggregated data. Such correlation involves identifying associations between data. For example, data A is ‘similar to’ or the ‘same as’ data B.
The query resolution function 27 can be implemented, e.g., in a stand-alone module, in a device 20 such as a set-top box or in a CE device 30 such as a DTV. An example implementation of extracting, aggregating and correlating data by the query resolution function 27 utilizing query plans is described below. XML-based execution plans are provided which encapsulate the steps involved in a search query resolution process. An execution plan comprises one or more plan-steps and each plan-step essentially specifies the type of task (i.e., data extraction, aggregation or correlation) to be performed.
Further, special classes, termed RuleLets, are provided to execute the three tasks (i.e., data extraction, aggregation or correlation) in a typical query resolution process. The RuleLets are: GetDataRuleLet, MergeDataRuleLet and GetContentNotInHomeRuleLet. The GetDataRuleLet obtains data from different data sources, the MergeDataRuleLet merges data obtained from different data sources and the GetContentNotInHomeRuleLet identifies the data/content (from a collection of data extracted from different sources) that are not available on the home devices.
A plan-step essentially specifies the RuleLet to be executed and the set of input and output parameters required for the execution of the RuleLet. The specific fields in a plan-step include the name of the RuleLet to be executed, the input data required for the RuleLet execution, the output-type expected from the execution of the RuleLet and the scope of the desired output data (if applicable). The scope field is used to specify whether the required data should be available in the home (“Local”) or on the “Internet.” In order to cater to different kinds of search queries, a plan library containing different kinds of plans is maintained. When a user chooses a search query, the query resolution function 27 identifies a plan based on the context of the user (e.g., if the user is watching a TV program, DVD or music video, or listening to a music album).
The use of execution plans in a search scenario in conjunction with example execution plans is described below. The search scenario involves a case where a user is watching a broadcast documentary program titled “Drumming Techniques” on a TV. When the user expresses interest to access related Internet content, the search facilitator 24 identifies and displays potential search queries from the program's closed captions (using the techniques described above) by executing the following plan steps: obtain the EPG related to the TV program being watched by the user; obtain keywords from the EPG information obtained in the previous step; obtain the genre of the TV program; based on the genre obtain significant keywords from the closed captions of the TV program; and merge the keywords identified from the EPG and the closed captions. An XML version of such a plan comprises:
The keywords obtained by executing this plan are then displayed to the user. One of the keywords/potential search queries displayed is: “Polyrthymic Drumming”. The user chooses “Polyrthymic Drumming” and expresses interest to see more related videos that the user has not seen before. To resolve this request, the facilitator 24 executes a plan, with “Polyrthymic Drumming” set as the keyword, including the plan steps: obtain videos related to the keyword (“Polyrthymic Drumming”) that are available on the Internet sources 66 (
The related Internet videos that are not already available in the local sources 69 are displayed to the user on the client module.
The CSF 82 includes a data and query processing (DQP) layer 83. The DQP 83 assists in resolving user queries and also provides an API for client applications 64 to make use of. Though client applications 64 are shown external to the CSF 82, the client applications 64 can also be components of the CSF 82. The DQP 83 includes a query execution planner (QEP) 84 and an information source manager (ISM) 85. The CSF 82 further includes a data execution (DE) layer 86. The DE 86 includes a data extraction manager (DEM) 87 and multiple plug-ins 88.
The QEP 84 provides interfaces for client applications to search for and access locally available data (i.e., data stored on the devices 30 and/or 20) and related data available on the Internet. The QEP 84 maintains a plan library 89, containing a set of pre-defined execution plans that are used to resolve requests for data. The QEP 84 also maintains the RuleLet 90 classes that are executed as part of a plan. When the QEP 84 receives a query from a client application, the QEP 84 retrieves the relevant plan from its plan library 89 and executes it. During the plan execution, the QEP 84 gathers the information/content requested by the user using the plug-ins 88 in the data extraction layer 86 (via the ISM 85). The ISM 85 manages a directory containing details about the types of data each data extraction plug-in component could extract and the input data (if any) expected by the plug-ins 88 to do so. This allows the QEP 84 to identify the plug-in 88 that provides a specific type of data.
The DE 86 includes many plug-ins 88 for extracting content/information from local and Internet data sources. 81 Local data sources refer to, e.g., home devices. Internet data sources include seed sources (e.g., BarnesandNoble.com, YouTube.com) and Internet search engines (e.g., Google, Yahoo). The functionalities provided by the different plug-ins 88 include: (1) A web scraper plug-in allows extracting specific information from specific websites; (2) A content manager plug-in allows accessing media content stored on the home devices; (3) An Internet video search plug-in allows searching for and accessing video content on the Internet; (4) A closed caption analyzer plug-in allows analyzing and identifying keywords from TV captions; and, (5) An EPG plug-in allows obtaining the EPG information for TV programs.
The DE 86 manages the plug-ins 88 and allows new plug-ins 88 to be added or removed with minimal code changes and provides an application programming interface for the higher-level components to use the plug-ins.
As an example of a search facilitation process by the CSF 82, according to the present invention, wherein a TV viewer accesses the Internet is as follows A user Trisha is watching a TV program on her TV about “Drumming Techniques” and is intrigued by the video. She wishes to learn more about the topics discussed in the program, especially about “Polyrhythmic drumming” which has just been mentioned. She presses a button on her TV remote control 31 and finds a host of information regarding the program being watched. A UI graphic on the client module screen shows two menus. One menu 64A provides a list of keywords related to the TV program (assembled by the query identification function of the CSF 82), and the first keyword “Polyrhythmic Drumming” is highlighted. The other menu 64B shows a list of search results (assembled by the query resolution function of the CSF 82) including web links containing information and/or videos related to the keyword “Polyrhythmic Drumming”. Trisha notices that the second link on this menu is a “how to” video. Using the navigation buttons on her remote control she highlights this link, and then presses the “enter” button to select the video and start viewing it.
The above scenario illustrates the following essential features: first, the user need not enter text or queries at any point; interaction is via the navigation buttons on a conventional remote control. Second, the user is able to access desired related Internet information by pushing a few buttons, as there is no need to bring up a search page or enter search terms. In this scenario, the context of the user (the program being watched), helps focus the search to relevant content.
The process for providing relevant information to a user of a CE device on a local network such as a home network generally involves:
Identifying correlations can be performed in one or more of the following example ways: (1) identifying correlations between information about current user activity and the interrelated information obtained from local sources, (2) identifying correlations between information about current user activity and the interrelated information obtained from external sources, and (3) identifying correlations between information about current user activity and the interrelated information obtained from local and external sources.
In order to minimize the number of keystrokes a user has to enter to receive information related to the current user activity, functionalities that support information searching are mapped to a small number of keys (e.g., mapping searches to a few keys of a remote control). Then, certain information is gathered about current user activity on CE devices. This includes obtaining metadata contained in media that is accessible only by content-rendering CE devices (e.g., the length and type of content contained in a CD or a DVD).
The process further involves obtaining information embedded in broadcast streams that are accessible only by a receiving/rendering CE device (e.g., subtitles and closed captions). In addition, information is gathered about content already existing on the home network (e.g., songs by Sting that are already owned by the user and the corresponding metadata). Further information is gathered about relevant structured data that exists on the Internet (e.g., gathering metadata about the songs already owned by the user from a compact disk database (CDDB)). Additional relevant information is obtained from semi-structured data that exists on the Internet (e.g., the biography of an artist from the Internet Movie Database (IMDb) and/or from the relevant web pages). Further relevant information is gathered from unstructured data that exists on the Internet (e.g., URLs of the web pages carrying the geographical, economical, political and cultural information about the place from which main events are being reported in the news).
The gathered/obtained information defines the information at hand. Then, when a user operates a CE device, what the user inputs to a CE device is correlated with the information at hand to automatically form queries to search for related information. This minimizes the need for the user to generate queries or use a keyboard in forming queries.
Then, from the information at hand, the data extracted from the Internet sources is correlated with the data extracted from home network content to form a query plan to refine the queries for precise searching. The query plan is then executed for searching the queries on the external network (e.g., the Internet, other resources), without requiring user intervention. The query execution results, in the form of search results, are then presented to the user. Preferably, based on the information at hand, the most relevant information from the search results is selected for presentation to the user, without requiring user intervention. Therefore, the information presented to the user includes information of potential interest to the user as related to the information at hand.
Another example of facilitating searches for the user involves obtaining information about current user activity on a local network, obtaining contextual information about current user activity on the local network, obtaining additional information interrelated to the contextual information and the user activity information, identifying correlations between the additional information, the contextual information and the user activity information, and using the correlations in forming a query to search for information related to the current user activity.
Obtaining additional information may include obtaining additional information interrelated to the contextual information and the user activity information, from sources including the local network and/or external sources. Identifying correlations may include identifying correlations between information about current user activity and interrelated information obtained from local sources. Identifying correlations may include identifying correlations between information about current user activity and the interrelated information obtained from external sources. Identifying correlations may include identifying correlations between information about current user activity and the interrelated information obtained from local and external sources.
Forming a query includes automatically forming a query, without requiring user intervention. The query is executed to obtain search results including information related to the current user activity. Executing the query further may include executing the query to search for related information on the local network and/or external sources. The search results may be presented to the user at this stage on a user interface in a device such as a CE device.
Obtaining information about current user activity on the local network may include obtaining information from user input to the device, or obtaining information from applications running in the network. Obtaining additional information may include obtaining the additional information from external structured data sources. Obtaining additional information may include obtaining additional information that is relevant to user interests from local media content sources.
Obtaining additional information may include obtaining the additional information from external unstructured data sources, from external semi-structured data sources, or from external broadcast data sources.
Obtaining contextual information about current user activity on the local network may include obtaining associated metadata available on the local network. As such forming a query may include using metadata related to the content on the local network for determining a context for query formation. Further, determining a context for query formation may include using metadata related to the content in the network and information from applications on the local network, to determine a context for query formation without requiring user intervention. The query may be used to search the Internet for information related to the current user activity or interest. As such, the above processes also enable improved access to the Internet to the users of CE devices.
The query identification function 25 (
Asynchronous resolution of queries may be beneficial for at least two reasons. First, if requested data is not available at the time of the query, the requested data can be provided to the user whenever it becomes available (without requiring the user to re-submit the query). Second, if the requested data is available at the time of the query, not only can the requested data be provided to the user immediately, but also continual data updates may be provided to the user as a background process. As such, the user is shown up to date data relevant to the user query.
According to the QEP 84 (
The plug-ins 88 in the data extraction layer 87 (
Upon receiving a query from the client, the QEP 84 (
The QEP 96 then executes the selected plan step-by-step, while passing on the generated query-id for the corresponding query to the plug-ins 99 (if specified in the plan step to do so). In response, the plug-ins 99 return query results (e.g., new data or data updates to the data sent initially) using events, along with said query-id, via the eventing module 93. The QEP 96 then checks the selected plan (e.g., look-up using the plan-id associated with the query-id), to determine if there are any plan steps that need to be re-executed because of the updates/new data returned by the plug-ins 99, and executes any such plan steps accordingly.
Further, the QEP 96 maintains a user profile in the user profile module 91, wherein the user profile comprises the preferences of the user (e.g., user preference about if/when the user wishes to see new keywords on the UI 64A). The QEP 96 may use the user profile to customize/rank the query results generated by execution of the selected plan.
As is known to those skilled in the art, the aforementioned example architectures described above, according to the present invention, can be implemented in many ways, such as program instructions for execution by a processor, program product stored on a computer useable medium, computer implemented method, as logic circuits, as an application specific integrated circuit, as firmware, etc. The present invention has been described in considerable detail with reference to certain preferred versions thereof, however, other versions are possible. Therefore, the spirit and scope of the appended claims should not be limited to the description of the preferred versions contained herein.
This application is a continuation-in-part of U.S. patent application Ser. No. 11/969,778, filed on Jan. 4, 2008, which in turn claims priority from U.S. Provisional Patent Application Ser. No. 60/898,257 filed on Jan. 29, 2007, incorporated herein by reference.
| Number | Date | Country | |
|---|---|---|---|
| 60898257 | Jan 2007 | US |
| Number | Date | Country | |
|---|---|---|---|
| Parent | 11969778 | Jan 2008 | US |
| Child | 12056184 | US |
