[The first type of machine learning technology: ranking algorithm]
This article refers to the address: http://
Ranking can be said to be one of the most important machine learning applications on the web. Companies large and small build business models around rankings, for example, the results returned by query strings. Quora uses different ranking algorithms for different purposes and for different purposes. An interesting example is the answer ranking. Assuming there are several answers to a question, we are interested in how to sort them in descending order so that the "best" answer is at the top and the worst answer is at the end. Determining the correct ordering of answers to a question involves multiple features. To determine the order, we first need to determine how Quora defines a "good answer." To get this definition, it's a good idea to look at the post "What kind of answers Quora thinks is useful", which will mention the answer to "real", "reusable", "giving an explanation", " Good format and so on. Our machine learning algorithm implements a special machine learning ranking method that uses a variety of features in an attempt to encode multiple dimensions associated with the above abstract concepts. For example, we use features that describe writing quality information, as well as features that describe the interactions that the answer receives (such as praise, tread, and number of expansions). We also use features related to the answer author, such as his professionalism in the problem domain. In Quora, there are many other ranking apps, some of which are not even noticeable. For example, a user name that is praised for an answer is also displayed after sorting, in order to rank the users we think are most knowledgeable about the question/answer. Similarly, when a possible answer is displayed for a particular question, those recommended users are also ranked. Let's take a closer look at two special cases of machine learning ranking algorithms: search and personalized ranking. [Machine Learning Ranking Algorithm Special Case 1: Search Algorithm] For applications such as Quora, the search algorithm can be considered as another application for ranking. In fact, search can be broken down into two steps: text matching and ranking. The first step is to somehow return the document (problem) that matches the query string entered in the search box. These documents are then ranked as candidates for the second step, so that the click probability and other aspects are optimized. Many of the features in the second step can be used, and it is indeed another example of a machine learning ranking algorithm. Includes simple text features that have been used during the initial text matching phase, as well as other features related to user behavior, or object properties such as popularity. [Machine Learning Ranking Algorithm Special Case 1: Personalized Ranking] In some scenarios as described above, perhaps a global optimal ranking for all users is sufficient. In other words, we can assume that the ordering of the most "helpful" answers for a given question is independent for the user reading the answer. However, this assumption does not hold true in many important situations. One of the occasions is the Quora Feed, which is basically a home page visible to users who logged into the product. On this homepage, we try to pick the most "fun" stories for a particular user at a particular time and rank them (see example below). This is a typical machine learning personalized ranking, similar to the Netflix home page for movies and TV shows. Quora's use cases are more challenging than Netflix movies and TV shows. In fact, our use cases can be seen as a combination of Netflix, Facebook, and Google News to optimize personalized rankings. On the one hand, we want to ensure that the top stories are relevant to the user on the topic. On the other hand, there is a clear relationship between Quora and the user. Your behavior on "social networks" should also have an impact on rankings. Again, the story on Quora may sometimes be associated with an ongoing trend event. Timeliness is another factor that should influence model decisions to determine whether a story should be ranked higher or lower. Because of this, Quora's personalized rankings involve a variety of different characteristics. Listed below are: 1. The quality of the question/answer; 2. The topic of interest to the user; 3. Other users that the user is interested in; 4. Popular events... In fact, it is important to remember that in Quora we are not only interested in attracting users. Interested in reading interesting content, and interested in submitting questions to users who can write interesting content. Therefore, we must incorporate features that are interesting to the answer and features that are specific to the problem. To get these features, we use information derived from users, authors, and object (such as answer/question) behavior. These behaviors are taken into account and accumulated over different time windows and provided to the ranking algorithm. In fact, we can get a lot of different features to join our personalized push model, and we have been trying to add more features. Another important consideration for our feed ranking app is that we need to be able to make millions of questions and answers about our usersâ€™ behaviors, perceptions, and even hot events, so we canâ€™t try to A user performs real-time rankings. To optimize the experience, we implemented a multi-segment ranking solution in which candidates were selected and sorted in advance, and then the final ranking was actually performed.
[Second type of machine learning technology: recommendation algorithm]
The above personalized ranking is already a form of recommendation. A similar approach is used in different cases. For example, the popular Quora Mail Collection includes a series of stories that are selected and recommended for you. This is a different machine learning ranking model that is optimized for different objective functions. In addition to the ranking algorithm, we have other personalized recommendation algorithms in different parts of the product. For example, in several places, you can see recommendations for people or topics. [Recommended basis: related questions] Another recommended source is to show users other issues that have some relationship with the current problem. The related problem is determined by another machine learning model that considers a number of different features, such as text similarity, co-visit data, or the same features as the subject. Features related to popularity, or quality of questions, should also be considered. It is necessary to point out that a good "similar problem" recommendation is not only how similar an item is to the source question, but also the "fun" of the target question. In fact, the most troublesome issue for any "related item" machine learning model is the trade-off between similarity and other relevance factors. Related Issues This model is especially effective for attracting logout users to access problem pages from external searches. This is one of the reasons why this recommendation model has not been personalized so far. [Recommended extremes: Duplicate problems] Duplicate problems are extreme cases of the above related problems. For Quora, this is a problem because we want to ensure that the user's energy to answer a specific question is shared and concentrated in the right place. Again, it is necessary to point out the answers already available to users who want to ask questions on the website. So, we spent a lot of effort to detect duplicate problems, especially at the stage of launching the problem. Our existing solution is based on a binary classifier that uses repeat/non-repetitive tag training. We use a variety of semaphores, from text vector space models to usage-based features.
[The third type of machine learning technology: user credibility / professional inference]
In an application like Quora, it is very important to grasp the credibility of users. In fact, we are not only totally limited to answering the question itself, but also interested in its relevance to related topics. A user may be knowledgeable about certain topics, but not necessarily in other areas. Quora uses machine learning techniques to infer user expertise. Not only do we understand what users have written for a given topic, but we also know how many praises, how many steps, and what comments. We also know how many "recommendations" this user has received in this area. Endorsements are a very clear recognition of a person's professionalism from the perspective of other users. There is another thing to keep in mind that credibility/professionality is spread through the network, which also needs to be considered by the algorithm. For example, if a machine learning expert gives a compliment to my answer in the field of machine learning, its weight should exceed the praise given by random users of non-experts in the field. This also applies to recommendations and other inter-user features.
[Fourth type of machine learning technology: spam detection and moderation]
Sites like Quora that are proud of maintaining high content quality must be wary of fooling the system with junk, malicious or very low quality content. Pure manual review mode cannot be extended. The solution to the problem, as you have guessed, is to use machine learning models to detect these problems. Quora has several models to detect content quality related issues. The output of these classifiers is not used directly for decision making in most cases, but instead provides these questions/answers to the throttle queue and then manually reviews them.
[The fifth type of machine learning technology: prediction of content creation]
For Quora, it's important to remember that we optimize many parts of the system, not only to attract readers, but also to produce the best quality, most popular content. Therefore, we have a machine learning model to predict the likelihood that a user will write an answer to a question. This allows our system to give priority to these issues in a variety of ways. One of them is that the system's automatic A2A (Ask to Answer) question is sent to potential respondents via prompts. The other ranking systems described above also use this model to predict probabilities.
[Core technical solution: establish an effective and flexible model]
Quora has tried many different models for the different cases described above. Sometimes we use open source implementations, but more often we end up with a more efficient and flexible build. I will not discuss the details of the model, but will list the models used by our system: 1, logistic regression; 2, elastic network; 3, gradient-enhanced decision tree; 4, random forest; 5, neural network; 6, LambdaMART; , matrix decomposition; 8, vector model and other natural language processing technology investment
In summary, Quora uses machine learning in a variety of ways. We have made very significant gains from using these machine learning methods, and we believe that there will be more revenue in the future, and we will continue to invest in new technologies. In addition, there will be exciting new machine learning applications in the near future, and we have already thought about it. These new applications, including ad rankings, machine translations, and other natural language processing areas, will directly become new features of the product we plan to add right away.
The handheld addresser is used to program the address of the monitoring module offline. When in use, connect the two output wires of the handheld encoder to the communication bus terminal (terminal label 1, 2) of the monitoring module, turn on the black power switch on the right side upwards, and press "ten Add", [Subtract ten", [Add one place" and [Subtract one place" to program the address.
Position Encoder,Magnetic Rotary Encoder,Hybrid Encoder,Channel Encoder
Changchun Guangxing Sensing Technology Co.LTD , https://www.gx-encoder.com