Patent Application
20150254213
The present method and user interface relates to methods of automatically summarizing content on a webpage, and, more particularly, automatically determining significant portions of text within an article or other long-form writing or data.
News and information articles can cover a wide variety of topics, and may solely exist on-line or have a corresponding print version, such as a newspaper or other periodical. The average article includes a headline and a body, where the body of a “long read” article has an average of 1000 words. When viewing a digitally delivered article on a website, RSS feed, or other digital delivery and viewing means, many readers read just the headlines. Often, those readers who begin reading the body of the article will not scroll down below the fold or load a second page to read the remainder of the article. Thus, the sole source of information for a great many readers is the headline and first few sentences. This leaves a substantial portion of the information contained in the body unread, leaving the average reader uninformed.
Whether in print or digital, the bodies of articles generally give background on the story, story context, interview quotes, and must provide a complete narrative by assuming the reader may know little about the subject matter. For stories that are ongoing, part of a series, or that cover developing situations, a large portion of the body may be dedicated to retelling prior versions of the story and filling in background, since many reader may not have read the previous related story. This repetitious model of storytelling creates overly-long and, for many, unreadable articles. Further, readers seldom have an hour to carefully read the day's articles in their entirety.
What is needed is a method and means for distilling the important aspects of a story into readable portions. The method and means should eliminate extraneous information that would cause an average reader to stop reading. The readable portion should be primarily contained above the fold, so that little or no scrolling is required.
The present system is provided for of analyzing text where the text comprises a one or more characters, the method comprising the steps of, under control of one or more computing systems configured with executable instructions, receiving a body of text, analyzing the body of text to detect a headline indicator for distinguishing a headline portion of the body of text; analyzing the body of text to detect a lead paragraph indicator for distinguishing a lead portion of the body of text, analyzing the body of the text to detect a conclusion paragraph indicator for distinguishing a conclusion portion of the body of text, and displaying the headline portion, the lead portion, and the conclusion portion within a graphical user interface.
Optionally, the headline indicator is one or more of a title tag, a headline tag, a headline portion location within the body of the text, a font size, and a font color. And optionally, the lead paragraph indicator is one or more of a sub-headline tag, a headline tag, and a lead portion within the body of the text relative to the headline portion. Optionally, a suspect lead portion may be excluded, if one or both of a word count and a character count is less than a minimum count in the suspect lead portion. One or both of the word count and the character count may be restricted to counting one or both of words and characters between a start tag and an end tag. The start tag may be one of a paragraph start tag and a heading start tag, and wherein the end tag is one of a paragraph end tag and a heading end tag. A suspect lead portion may be excluded, if the suspect lead portion contains text matching one or more of a list of excluded text.
As an option, the number of paragraph elements may be counted by the software to determine a total number of paragraphs in the body of text. And the position of each counted paragraph may be determined relative to the remaining paragraphs. A mid-portion of the body of text may be determined by finding the quotient of the total number of paragraphs divided by two. Text between heading elements may be excluded in counting the total number of paragraphs.
As yet another option, at least one of the headline indicator, the lead paragraph indicator, and the conclusion paragraph indicator may be at least one HTML element. The body of text may be received from one of an address on the World Wide Web, a local server, and a remote server. Optionally, the body of text may be displayed in a first window within the graphical user interface. Emphasis may be added to at least one of the headline portion, the lead portion, and the conclusion portion within the body of the text in the first window, such as highlighting, underlining, and bolding.
Further, as an option, at least one of the headline portion, the lead portion, and the conclusion portion in a second window within the graphical user interface may be displayed in isolation from the body of the text. Editing of at least one of the headline portion, the lead portion, and the conclusion portion may be permitted within the second window. The headline portion may be displayed in a third window within the graphical user interface. Editing of the headline portion may be permitted within the third window. An edited headline portion, an edited lead portion, and an edited conclusion portion may be displayed in a fourth window within the graphical user interface.
Optionally, an image search may be initiated using selected keywords found within at least one of the first window, the second window, the third window, the fourth window, a keyword metadata, a summary metadata, a title tag, and a heading tag. At least one image may be associated with the edited headline portion, the edited lead portion, and the edited conclusion portion in a fourth window, where the image may be selected by an editor or automatically found within the image search query.
The edited headline portion, the edited lead portion, the edited conclusion portion, and the image may be optionally published within a reader user interface, such that a user may read the edited text and view the images together.
Additional objects and features if the method will be more readily apparent from the following detailed description and appended claims when taken in conjunction with the drawings, in which:
The detailed descriptions set forth below in connection with the appended drawings are intended as a description of embodiments of the invention, and is not intended to represent the only forms in which the present invention may be constructed and/or utilized. The descriptions set forth the structure and the sequence of steps for constructing and operating the invention in connection with the illustrated embodiments. It is to be understood, however, that the same or equivalent structures and steps may be accomplished by different embodiments that are also intended to be encompassed within the spirit and scope of the invention.
The present system and method provide a user interface tool for distilling a long-form article into one or more bullet points, with each bullet point having an associated image displayed in proximity to the bullet point, so that the reader quickly understands the primary aspects of a story through reading the text and viewing the associated image.
Example computer networks are well known in the art, often having one or more client computers and a server, on which any of the methods and systems of various embodiments may be implemented. In particular the computer system, or server in this example, may represent any of the computer systems and physical components necessary to perform the computerized methods discussed in connection with
The illustrated exemplary server and client computer are known to a person of ordinary skill in the art, and may include a processor, a bus for communicating information, a main memory coupled to the bus for storing information and instructions to be executed by the processor and for storing temporary variables or other intermediate information during the execution of instructions by processor, a static storage device or other non-transitory computer readable medium for storing static information and instructions for the processor, and a storage device, such as a hard disk, may also be provided and coupled to the bus for storing information and instructions. The server and client computers may optionally be coupled to a display for displaying information. However, in the case of servers, such a display may not be present and all administration of the server may be via remote clients. Further, the server and client computers may optionally include an input device for communicating information and command selections to the processor, such as a keyboard, mouse, touchpad, and the like.
The server and client computers may also include a communication interface coupled to the bus, for providing two-way, wired and/or wireless data communication to and from the server and/or client computers. For example, the communications interface may send and receive signals via a local area network or other network, including the Internet.
In the present illustrated example, the hard drive of the server or the client computer is encoded with executable instructions, that when executed by a processor cause the processor to perform acts as described in the methods of
An exemplary long-form article (200) is schematically illustrated in
An original article midpoint (212) general location is also indicated. The midpoint (212) can be determined automatically though an executable program that counts the total number of paragraphs in either the story body (208) or the entire article (200) and divides that number by two to determine the approximate midpoint paragraph number, where the paragraph numbering may start from the first full paragraph at the top of the article. For example, the executable program may count the number of start tag (<p>) and end tag (</p>) pairs between main element pairs (<main> and </main>), div element pairs (<div> and </div>), or other indicators of the start and end of the article. Then, the executable program (software) seeks the paragraph or paragraphs at or near the midpoint number of paragraphs. The midpoint is selected usually because important information may be located at or near the midpoint (212). Of course, if a pattern is discovered which locates critical information in another general area (e.g., one-third down, two thirds down, etc.), then the algorithm of the executable program can be adjusted to locate paragraphs in that general location in the article (200).
The executable software can also be used to determine the location and text of the headline (202), by detecting headline indicators, such as location, formatting, font size, font color, and other factors usually associated with headlines in general. These headline indicators can usually be found in the source code (HTML, etc.), such as a title tag (<title>XXXX(</title>) which would be displayed in the browser's title bar, a heading tag (<h1>XXXX</h1> or <h1 class=“title” itemprop=“headline”>XXXX</h1>), or similar indicator (like <hgroup> or similar); the series of X's represent text within the article headline or title. Generally, the headline (202) is located at or near the top of the article (200). Also, generally, the headline (202) font size is larger than the remainder of the article (202). Once the software has determined the text most likely to be a headline, that text can be highlighted, labeled, and/or classified as a headline (202). If other heading elements are present (such as <h2>, <h3>, <h4>, <h5>, or <h6>), the elements may be ranked according to importance, where <h2> is most important after <h1> and <h6> is least important. Although the exemplary code is HTML, any code for building an article page within a browser or similar display means may be analyzed and classified in a similar manner.
Often, an article (200) may have lead (204) or sub-headline, which is one or more short sentences or a sentence fragment at or near the top of the article (200) that piques the interest of the reader and causes her to become interested in the article (200). Like the headline (202), the lead (204) can be determined by the software by various indicators, such as a sub-headline tag or element (such as <h2 class=“sub-head” itemprop=“description”>XXXX</h2> or similar). The lead (204) is generally just beneath the headline (202). However, other non-essential information may also be in this location, such as the author's name, the date, the news outlet, and other information not pertinent to the story. Thus, certain keywords may be sought out, such as a line beginning with “by” or other keyword indicative of an author's name or a known news outlet (or elements, such as <address>). Further, the software may count the number of words and exclude any paragraph or sentence fragment with a word count that falls below a minimal threshold. For example, the software may exclude isolated paragraphs or sentences just beneath the headline and having less than five words. In this way, non-pertinent information is often excluded, minimized, or merely not highlighted. The minimum word count can change, depending on the circumstances. Once the most likely lead paragraph (204) is determined, it is highlighted and/or classified as a lead paragraph (204). The analysis to determine the most likely lead paragraph (206) may be restricted to text between paragraph elements (<p>XXXX</p>). Thus, in this example, the number of words between the start tag <p> and the end tag </p> (or other tag indicating the end of the paragraph) for each paragraph may be counted.
The nutshell paragraph (206) is generally one or two paragraphs that explain why the story is important, by providing the theme of the story and supporting facts or information. Basically, the “who, what, when, where, and why” is most likely provided in the nutshell (206). The nutshell paragraph is often just beneath the lead paragraph (204); or if there is no lead paragraph (204), the nutshell may be just below the headline (202). The program can often determine the nutshell paragraphs, again, by looking at certain indicators. Since supporting information is often provided in the nutshell paragraph (206) (such as dates, numbers, names, locations, etc.), the software algorithm can be optimized to seek out numbers, known names of public or private figures, words in mid-sentence starting with or having a capital letter, words preceding “Inc.”, and other indicators of important facts. Once a paragraph or two adjoining paragraphs are discovered meeting one or more of the above criteria, then that paragraph(s) is highlighted and/or classified as a nutshell paragraph. Yet another indicator of a nutshell paragraph may be determined by analyzing the metadata, such as the article description or summary metadata <meta name=“article.summary” content=“XXXXX.”/>. Further keywords from the keyword metadata may be used to search the article for matching keywords and/or a high density of matching keywords to determine the most important paragraphs, the nutshell paragraphs, or other paragraphs of interest.
The conclusion (210) is most often found at the very end of the article (200), at the last paragraph. Thus, the last paragraph that meets the minimum word count, will be highlighted and/or classified as a conclusion paragraph (210). As above, paragraph elements (<p> and/or </p>) or div elements (<div> and/or </div>) may be used to determine the final paragraph. Additionally, other elements may be used to indicate the final paragraph, such as the footer element (<footer>) or other indicator that the article text has ended. For example, the div or footer elements may indicate that the article ended one or more lines (of code) above the div or footer element, such as the closest prior </p> element or other end tag.
Paragraph-by-paragraph classification of the original article (200) may be completed automatically using the above described filtering criteria. This classification generally occurs when the URL is called up and built by the present software. Additionally, a list of URL's may be automatically generated, so that the software downloads the website associated with the URL and classifies the article (200).
The editor may manually enter the text into the URL entry box (70), which will cause the article (200) associated with the URL to be downloaded and classified as described above. Then, the article headline is displayed within the list of article headlines (76, 78, 80). In the illustrated example embodiment, there are three article headlines in the list of article headlines (76, 78, 80); however, this list may include more or less headlines.
Selecting the edit article icon (82) opens and displays the bulleting tool user interface (20) for the article associated with that particular edit article icon (82) (e.g., the icon button may be aligned or within the same area as the headline for that article), as schematically illustrated in
The human editor has the option of reading the entire article (200) or just the automatically highlighted portions. The human editor can deselect an automatically highlighted portion, if she believes the portion is not pertinent to the story. The human editor can also select further portions by right-clicking and “mousing” over the desired text portion to create additional highlighted portions. Upon releasing the right mouse button, a confirmation box may be displayed which queries if the editor desires to add the user-highlighted portion to the highlighted portions box (24). If the editor confirms, then the highlighted portion, in its entirety, is moved to the highlighted portions box (24). As is well known in word processing, the selected text may also be moved by selecting with a mouse gesture, then clicking and dragging the text to the highlighted portions box (24).
The human editor also has the option of skipping the article listing user interface (86), and directly entering the text of a URL address into the URL entry box (70) within the bulleting tool user interface (20). The article associated with the URL is called up, classified, and then displayed as text in the article box (22). In this way, the human editor has the option of completely overriding the automatic classification of the article text. However, the human editing may solely be based on the displayed text of the article, and not the HTML code. Thus, for more complex articles or articles written in a non-standard format (perhaps a machine-translated article), a human editor may be required to refine the automatic classification.
Much like the article listing user interface (86) of
Just to the right of the original article box (22) is the first summary area (25), with the headline box (26) and the highlighted portions box (24). The headline box (26) displays the headline (27) as editable text. When the human editor first views the headline box (26), the box (26) may be empty or the portion of the original article (22) that is determined by the software to most likely be the headline is automatically placed as text into the headline box (26). The human editor may edit the text or select new text from the article (200) to replace the automatically populated text. For example, a long headline may be shortened or changed completely.
Below the headline box (26) is the highlighted portions box (24), which would include all of the highlighted and/or selected portions of the original article (200). The highlighted portions box (24) may be initially empty or may be automatically populated with the portions selected by the software. For example, the first selected portion (28) may be the text of the lead (204) or nutshell (206) paragraphs, the second selected portion (30) may be the text of the midpoint paragraph (212), and the third selected portion may be the conclusion paragraph (210). Of course, there may be more or less than three selected portions, depending on the story and the editor's preference. Next to each selected portion (28, 30, 32) is a delete icon (34), which will remove the selected portion (28, 30, 32) associated with the delete icon (34) once selected. In this way, the human editor can add or remove text from the original article (200) to or from the highlighted portions box (24). Thus, the highlighted portions box (24) enables a preliminary round of distilling, where the text from entire selected portions of the original article (200) are displayed in the highlighted portions box (24), with each separate highlighted portion from the original article (200) displayed as a separate selected portion (28, 30, or 32).
To the right of the highlighted portions box (24) is the final summary area (36), where the human editor creates final bullet points (56, 58, 60). The number of bullet point boxes (42, 44, 46) are determined either manually by the editor, automatically by the number of selected portions (28, 30, 32), or may be a fixed number of boxes. The human editor either manually enters the text into each bullet point box (42, 44, 46) or copies parts of the text from the selected portions (28, 30, 32). The goal is to further summarize the information from the highlighted portions box (24) to create several final bullet points (56, 58, 60). The operation of creating final bullet points (56, 58, 60) may be automatically achieved through software analysis of the selected portions (28, 30, or 32). The software may select pertinent words, indicating names, dates, quantities, and so on, to form short sentences or fragments that can serve as final bullet points (56, 58, 60).
Next to each bullet point box (42, 44, 46) is an associate image (48, 50, 52). For example, looking at bullet point box (42), when the bullet point box (42) is empty, there is no associated image (48). When the editor enters the text of the final bullet point (56), the text is submitted to a search engine, which conducts an image search based on the text or selected portion of the text in the final bullet point (56). The image search generally produces multiple images, from which the editor may select the image which most closely conveys the subject matter of the associate bullet point (56) or other bullet point (58 or 60). Once a first associated image (48) is selected, it is displayed as a thumbnail. The second and third associated images (50, 52) may or may not be selected. If selected, the second and third associated images (50, 52) are generally complementary to the first associated image (48), by furthering the story communicated with the bullet points (56, 58, 60) and providing visual information differing from the other associated images. Additionally, the metadata on the site page may be used in determining the associate image (48, 50, 52). For example, the “keywords” or “news_keywords” metadata may be used to determine the image search keywords. In another example image links may be designated by the site within the metadata for use with social media or other quoting source, such as <meta name=“twitter:image” content=“http://website.com/images/samplepic.jpg”/> or <meta property=“og:image” content=“http://website.com/images/samplepic.jpg”/>. The image link designated in the metadata are generally closely related to the story, as they were selected by the original publisher.
Since there is a strong desire to maintain brevity in the final bullet points (56, 58, 60), a word count (40) and a character count (38) may be provided and limited. The character and word limits set may be inextensible or may merely provide an alert to the editor that the total characters/words of all the final bullet points (56, 58, 60) combined exceed the recommended limit.
Once the final bullet points (56, 58, 60) are complete and the associated images (48, 50, 52) are selected, the editor may select the save article icon (64) to save the progress and open the article listing user interface (86) of
In particular,
The headline (27) text may displayed on top of the image pane (100) in large text. The top portion of the image pane (100) may have a gradient effect to darkly shade the top portion, so that the headline (27) is more prominently displayed. In the reading pane (98), the final bullet points (56, 58, 60) are displayed in three distinct sentences or fragments, so that the reader may easily read and understand the bullet points. If more or less than three final bullet points (56, 58, 60), then the number of bullet points (56, 58, 60) in the reading pane (98) will be similarly adjusted.
The individually created story summaries are displayed sequentially, much like a slide show, retrieved form a list of completed and published summaries. The reader has the option of returning to a previously displayed story by selecting the previous story icon (90), or skipping to the next story by selecting the next story icon (96). The reader may select the pause icon (92) or the play icon (94) to continue. Further, the reader may select the reading pane (98) or the image pane (100) to be directed to the original article (200) or the source of the associated images (48, 50, 52). Alternatively or in addition to selecting icons to navigate to the next story, the reader user interface (88) may be configured to display the next story by receiving a swiping input, such as from a touch screen device or mouse action.
If there is an ongoing story that requires several summaries over time based on several articles, the software can optionally conduct a comparative analysis of the text of the related articles to determine which portions of the related stories are new and which are portions are repetitive of the prior story. For example, a first article in a series of articles may extensively explain the background of the story. A second article in the series may still include much of the background from the first article to fill in readers who missed the first. The comparative analysis would eliminate parts of the second article which substantially repeat the first article, by looking at similarities of groupings of words within a sentence and comparing to similar sentences in the first article, or detecting repetitive quantitative facts, and so on. In this way, the summaries in which summarize several related articles only includes new information.
By present system and method employs distributed learning by associating a bullet point to an image that emphasizes the information provided in the bullet point. This engages multiple parts of the reader's brain. Further, by inducing motion within the image pane, such as pushing or pulling the image or panning, the reader is more engaged, resulting in higher retention of information.
This application claims the priority date of provisional application No. 61/939,226 filed on Feb. 12, 2014, which is hereby incorporated by reference in its entirety.
| Number | Date | Country | |
|---|---|---|---|
| 61939226 | Feb 2014 | US |
